System and Method for Assessing Sleep State

Abstract

Assessing sleep state of an individual. A time series of accelerometer data is obtained from an accelerometer device mounted upon or to the individual. From the time series of accelerometer data a percentage of time in which the individual is substantially immobile (% TA) is determined. From the time series of accelerometer data a typical time of continuous immobility (MTI) is also determined. The % TA and MTI are combined such as by weighted sum, to produce a sleep score. If the sleep score exceeds a threshold, this is an indication that the individual is asleep.

Claims

1. A method of assessing sleep state of an individual, the method comprising: obtaining a time series of accelerometer data from an accelerometer device mounted upon or to the individual; determining from the time series of accelerometer data a percentage of time in which the individual is substantially immobile (% TA); determining from the time series of accelerometer data a typical time of continuous immobility (MTI); combining the % TA and MTI to produce a sleep score; and if the sleep score exceeds a threshold, outputting an indication that the individual is asleep.

2. The method of claim 1 when used in a non-clinical setting such as the individual's home.

3. The method of claim 1 applied to assess sleep state of Parkinsonian subjects.

4. The method of claim 1 applied to assess sleep state of non-Parkinsonian subjects.

5. The method of claim 1 wherein producing the sleep score further comprises summing or otherwise combining 2 or more of a set of sleep-related variables derived from the accelerometer data.

6. The method of claim 5 wherein the sleep related variables include a variable reflecting the individual's attempts at being active.

7. The method of claim 5 wherein the sleep related variables include a variable reflecting the individual's inactivity while awake.

8. The method of claim 5 wherein the sleep related variables include a variable reflecting the individual's immobility while asleep.

9. The method of claim 5 wherein the sleep related variables include a variable reflecting the individual's sleep duration.

10. The method of claim 5 wherein the sleep related variables include a variable reflecting the individual's sleep fragment length.

11. The method of claim 5 wherein the sleep related variables include a variable reflecting the individual's Sleep Quality.

12. A system for assessing sleep state of an individual, the system comprising: an accelerometer device configured to be mounted upon or to the individual and configured to obtain a time series of accelerometer data; and a processor configured to determine from the time series of accelerometer data a percentage of time in which the individual is substantially immobile (% TA), the processor further configured to determine from the time series of accelerometer data a typical time of continuous immobility (MTI); the processor further configured to combine the % TA and MTI to produce a sleep score; and the processor further configured to, if the sleep score exceeds a threshold, output an indication that the individual is asleep.

13. A non-transitory computer readable medium for assessing sleep state of an individual, comprising instructions which, when executed by one or more processors, causes performance of the following: obtaining a time series of accelerometer data from an accelerometer device mounted upon or to the individual; determining from the time series of accelerometer data a percentage of time in which the individual is immobile (% TA); determining from the time series of accelerometer data a median or typical time of continuous immobility (MTI); combining the % TA and MTI to produce a sleep score; and if the sleep score exceeds a threshold, outputting an indication that the individual is asleep.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0036] An example of the invention will now be described with reference to the accompanying drawings, in which:

[0037] FIGS. 1-3 illustrate a means for detection of kinetic state in accordance with an embodiment of the invention;

[0038] FIGS. 4-6 illustrate the efficacy of the described approach.

[0039] FIGS. 7A-7E illustrate sleep state of a control subject;

[0040] FIGS. 8A-8C illustrate sleep state of another control subject;

[0041] FIGS. 9A-9C illustrate sleep state of a person with Parkinson's;

[0042] FIGS. 10A-10C illustrate sleep state of another person with Parkinson's;

[0043] FIGS. 11A-11H illustrate the relative statistical importance of sleep state variables in differentiating differing sleep states;

[0044] FIGS. 12A-12I illustrate the relative statistical importance of sleep state variables, and sleep state scores derived therefrom, in differentiating differing sleep states;

[0045] FIGS. 13A-13I illustrate the correlation of sleep state variables, and sleep state scores derived therefrom, to a clinical standard; and

[0046] FIGS. 14A-14I illustrate the relationship between each variable and the PSG score.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0047] FIG. 1 is a diagrammatic view of a device 15 for detection of kinetic state during an attempted sleep period of an individual, in accordance with an embodiment of the invention. The device 15 is wrist mounted which the present inventors have recognised provides a sufficiently accurate representation of the kinetic state of the whole body. The device 15 comprises three elements for obtaining movement data of a limb of a person. The device 15 comprises a motion monitor 21 in the form of an accelerometer, a data store 22 for recording the data, and an output means 23 for outputting movement data.

[0048] The device 15 is a light weight device which is intended to be worn on the wrist of the person as shown in FIG. 2. The device is mounted on an elastic wrist band so as to be firmly supported enough that it does not wobble on the arm and therefore does not exaggerate accelerations. The device is configured to rise away from the person's wrist by a minimal amount, or not at all, so as to minimise exaggeration of movements. The device may be on a wrist band secured by a buckle, whereby the act of unbuckling and removing the device breaks a circuit and informs the logger that the device is not being worn.

[0049] The user preferably wears the device throughout the night or throughout an attempted sleep period of interest. This allows the device to record kinetic activity of the individual for the sleep period. The accelerometer 21 records acceleration in three axes X, Y, Z over the bandwidth 0-10 Hz, and stores the three channels of data in memory on-board the device. This device has sufficient storage to allow data to be stored on the device for a recording period of up to 12 hours, more preferably 10 days, after which the device can be provided to an administrator for the data to be downloaded and analysed. Additionally, in this embodiment, when the device is removed after the recording period, the device is configured to transfer the data to an associated device which then transmits the data via wireless broadband to analysis servers at a central facility (114 in FIG. 3).

[0050] FIG. 3 illustrates kinetic state monitoring and reporting in accordance with one embodiment of the invention. A user 112 is wearing the device of FIGS. 1 & 2. The device 15 logs accelerometer data and communicates it to a central computing facility 114. The computing facility 114 analyses the data using an algorithm (discussed further below), to obtain a time series of scores for the sleep state of the person 112. These scores are reported to a sleep physician 116 in a format which can be rapidly interpreted by the sleep physician to ensure efficient use of the physician's time. Physician 116 then interprets the sleep state report and implements or updates a treatment of the user 112 as required.

[0051] The accelerometer 21 measures acceleration using a uniaxial accelerometer with a measurement range of +/4 g over a frequency range of 0 to 10 Hz. Alternatively a triaxial accelerometer can be used to provide greater sensitivity.

[0052] In this embodiment algorithms are applied to the obtained data by a central computing facility 114 in order to generate an assessment of a sleep state of the individual, referred to in the following as a PKG measure or score.

[0053] Method

[0054] In a first embodiment of the invention, described in relation to FIGS. 4 to 6, we performed simultaneous Polysomnography (PSG) and PKG measures in 45 subjects, 10 of whom had normal sleep. The PKG scores periods of immobility of at least 2 minutes and we used this to develop, amongst other measures, surrogates for SE (percent of attempted sleep time in which the patient was immobile) and fragmentation (the median length of each period of continuous immobility). These are called % time asleep (% TA) and median time immobile (MTI) respectively. We used these to develop a score that clearly differentiated between normal and abnormal sleepers as determined by the PSG report.

[0055] We then applied this PKG score to 24 age matched subjects without PD and 35 people with PD (PwP) who wore the PKG for 6 nights and responded to various questionnaires including Parkinson's Disease Sleep Scale 2 (PDSS-2). A further 45 PKG subjects were also analysed but without questionnaire.

[0056] Results

[0057] The PKG score combining the % TA and the MTI predicted normal or abnormal sleep (according to the PSG) with 100% selectivity and sensitivity. In the 24 subjects without PD only 2 had abnormal sleep according to the PKG and one of these gave a history of restless legs. Amongst the PD subjects 28% had normal sleep according to the PKG criteria and in those interviewed, PKG values had a good correlation (r2=0.49) with the PDSS2 scale.

[0058] Conclusions

[0059] The PKG score appears to provide a simple means of detecting normal and abnormal sleep in PD. This is based on a small PSG sample.

[0060] The above example is now described in further detail. The sources of patients studied were as follows: 36 from Monash (Victoria, Australia) sleep lab and 9 from an epilepsy study (none of whom were thought to have a sleep disorder.

[0061] Normal (N): 8 of the epilepsy patients and two of the Monash sleep patients were reported as having normal sleep. Sleep Disordered (SD): See Table 2 for PSG diagnosis (col 3), scores from PSG (Col 4-7) and our classification (Col 2), which was based on the PSG diagnosis as shown in Table 1.

TABLE-US-00001 TABLE 1 PSG Mild Severe Controls Mild 1 plus mild-mod Mod minus Severe Score 0 1 2 3 4 5 6

TABLE-US-00002 TABLE 2 (note, spread over 2 pages): 1) PKG 2) PSG 4) Sleep 5) Score Score 3) Conclusion efficiency PLM 6) AI 7) AHI 8 0 Unremarkable study 82.3 3.3 11.9 0.7 8 0 No significant sleep disordered 85.9 0 1.8 0.7 breathing 8 0 There is no significant sleep 93.2 6.8 12.1 2.3 disordered breathing MONASH 6 0 Normal sleep MONASH* 75.4 7.7 25.1 0.5 5 0 Normal study 84.9 0 19.1 0.5 7 0 Unremarkable study. Periodic leg 90.4 21.8 17.5 1.5 movements were present and infrequently associated with cortical arousal 8 0 No significant sleep disordered 88.5 27.5 11.2 1.5 breathing 8 0 no significant sleep disordered 84.3 0.2 18 0.8 breathing 7 0 Unremarkable study. PLM not 95.2 12.5 23.3 1 significant. 6 0 no significant sleep disordered 76 11.9 21.6 5.2 breathing 3 1 mild REM predominant sleep 82.5 7.1 23 30 disordered breathing with mild arterial oxygen desaturations in REM and stable SpO2 in NREM EPILEPSY STUDY 4 1 Mild REM based OSA 89.5 0 6.6 6.1 4 1 Normal? 60.9 0 6.8 0.8 4 1 Mild OSA 88.4 9.9 12.9 6.1 2 1 Adequate CPAP 60.1 7.4 7.4 8.6 3 1 Mild OSA 57.3 10.7 22.1 6.1 2 1 Normal study with high sleep 87 0 10.9 0.1 tendency 2 1 No OSA, no narcolepsy, fragmented 85.6 2.4 25.4 1.4 sleep 3 1 Mild OSA 84.6 0 11.6 6.6 2 2 Mild OSA, poor sleep efficiency 57 1.3 14.2 5.6 3 2 Baseline O2 sats 92 fell to 89% 77.9 0 11.8 1.6 when asleep 4 2 Fragmented sleep 81.5 0 8.9 0.2 3 2 Mild OSA, fragmented sleep 71.5 0.8 13.3 9.8 3 2 Mild OSA, clusters of PLM 78.3 13.3 18.2 9.7 4 2 Mean reduced sleep latency-severe 91.3 8.1 21.8 4.1 sleepiness 2 2 OSA, fragmentation 68 0 24 38.8 2 3 Mild-Mod OSA 86.4 2.3 22.7 22.7 2 3 Mod OSA 85.6 0 25.6 19.2 2 4 Mod OSA 69.1 21.2 21.6 6.5 4 4 Mod OSA 84.1 0 29.5 22.1 3 4 Mod OSA 86.5 4.3 36.8 21.4 3 4 Mod OSA 85.7 0 26.2 20.2 2 4 Mod-Severe OSA 87 58.3 13.8 26.9 3 5 Severe supine OSA (mild lateral) 90.6 0 62.9 58.1 4 5 Mod OSA, fragmented sleep 91.4 12.2 37.6 19.6 2 5 New CPAP levels prescribed 83.1 43.9 34.2 6.6 3 5 Mod-Severe OSA when supine 85 24.2 12.6 13.2 2 6 Severe OSA 67.1 0 5.8 18.6 3 6 Severe OSA 72.7 0 14.9 33.8 3 6 Poor sleep 66.6 32.6 15.6 3.6 3 6 Severe OSA 62.5 4.7 27.7 48.8 4 6 Severe OSA 73.8 0 40.5 32.8 3 6 Severe OSA 82.5 0 26.9 52.9 2 6 Severe OSA 91.9 0 26.3 57.6 2 6 Severe OSA, very abnormal sleep 48.3 25.4 45.3 38.7 2 6 Severe OSA, fragmented sleep 86.4 0 35.3 41.3 3 6 Severe OSA 91.2 0 44.1 89.4

[0062] Column 1 of Table 2 contains the score as estimated from the PKG measures in accordance with the present embodiment of the invention. Two values were used to produce a PKG score: The Percent Time Asleep, which is a measure of the proportion of time immobile over the period in which sleep was attempted (akin to sleep efficiency) and the median length (duration) of each period of immobility making up the sleep (akin to a measure of fragmentation). A number of other markers were examined but these two provided a degree of difference between SD and N. Percent Time Asleep was then scored with a level of severity from 1-5 (with 1 being most affected and 5 being normal) based on the median, 75th and 90th percentile of normals (for Percent Time Asleep) as well as the 75th percentile of SD. Median Duration of immobility was then scored with a level of severity from 1-3 (with 1 being most affected and 3 being normal) based on the median and 75th of normal.

[0063] The Scores for the PKG and the PSG were then compared (FIG. 4). A score of 1.5 for immobility gave a high sensitivity (90%) for finding normal subjects but with very poor selectivity (38%). A score of 3.5 for % time asleep gave a sensitivity (90%) for finding normal subjects with an improved selectivity (70%). Adding the two (as shown in FIG. 5) however gave complete separation between normal and abnormal subjects when the used threshold score was 4.5. This suggest that the combined PKG score would be a good screening tool to detect abnormal sleep if a score of <5 or even <6 was used. Combining the scores can be simple addition of the scores as shown in FIG. 5 or in other embodiments could be any other suitable linear or more complex non-linear combination of the % TA and MTI scores which elicits improved sensitivity and selectivity.

[0064] The next step was to compare the PKG score with the PSG (Table 1, FIG. 6). This further confirms that the PKG score is helpful for sorting into normal and abnormal sleep but not in grading severity further in terms of matching severity by PSG. Note that the sleep abnormalities in the PSG were most severe for OSA and these are not necessarily the reason for having abnormalities of sleep in PD.

[0065] FIGS. 7 to 14 illustrate further embodiments of the present invention. In these embodiments, normal ranges for the respective scores were obtained from a cohort of 155 subjects aged 60 years or more without known neurodegenerative disorders. The comparison group was 72 PD subjects.

[0066] The various scores assessed, and their derivation, is as follows. The time period of data recording was divided into periods based on the time of day, as follows. An Active Period (AP) during the hours 09:00-18:00, chosen because most subjects are active and pursuing their usual daily activity in this period. A Night Period (NP) was examined for quality of nocturnal sleep. A Rest Period (RP) during the hours 08:00-23:00 was chosen to represent a period when most people are sedentary.

[0067] Definitions of Movements

[0068] A dyskinesia score (DKS, or DK score) is calculated every two minutes throughout the period of time that the logger is worn. In the presently described embodiments the DKS is calculated in accordance with the teachings of International Patent Publication Number WO 2009/149520, the content of which is incorporated herein by reference, however in alternative embodiments the DKS may be determined in any suitable alternative manner.

[0069] Median DKS. The median value of the DK scores from the AP. The Median DKS correlates with the Abnormal Involuntary Movement Score assessed at the time of donning the PKG logger. FIG. 11b shows the Distribution of the median DKS for the control group and the PD Group. Table 3 below sets out the values observed for DKS in each group, in particular being the minimum observed DKS value, the 10.sup.th, 25.sup.th, 75.sup.th and 90.sup.th percentile values of DKS, the Median DKS, and the maximum observed DKS value. It is to be noted that DKS may be measured on any suitable scale, and may be assessed by reference to any suitable division of percentile bands. For example alternative embodiments of the present invention may use four percentile bands in the manner described in the above-referenced WO 2009/149520, specifically DK I (0-50th percentile of normal) DK II (50.sup.th-75.sup.th percentile of normal), DK III (75.sup.th-90.sup.th percentile of normal) and DK IV (>90.sup.th percentile of normal).

TABLE-US-00003 TABLE 3 SCORES THAT CONTRIBUTE TO AP ASSESSMENT PERCENTILE Min 10% 25% Median 75% 90% Max DKS C 0.2 0.66 1.2 2.1 3.6 6.22 18.1 PD 0.1 0.33 0.725 2.2 4.65 6.54 11.5 BKS.sub. C 13.2 17.82 20.5 22.6 24.9 28.04 31.4 PD 14.5 18.15 20.83 24 28.63 35.14 50.9 ACTIVE.sub.50 C 12 15.96 17.8 19.9 22.4 24.32 29.8 PD 13.8 15.7 18.2 21.05 24.98 31.13 47.7 Boundary.sub.A-I C 24 30 33 36 40 44.4 63 PD 27 30 32 38 44 51 65 PTA C 49.1 65.98 71.2 78.5 84.8 91.04 95.4 PD 42.6 56.09 68.03 76.15 84.28 89.95 92.2 PTIn C 1.4 7.52 10.6 16 22.2 28.24 41.6 PD 1.8 6.26 9.075 15.85 24.15 29.14 43.9 PTI C 0.1 1.2 2.1 4 6.9 10.24 18.9 PD 0.1 1.1 3.525 6.6 11.13 16.78 31.5

[0070] A bradykinesia score (BKS, or BK score) is calculated every two minutes throughout the period of time that the logger is worn. In the presently described embodiments each BKS is calculated in accordance with the teachings of International Patent Publication Number WO 2009/149520, the content of which is incorporated herein by reference, however in alternative embodiments the BKS may be determined in any suitable alternative manner. It is to be noted that, as for DKS, the BKS may be measured on any suitable scale, and may be assessed by reference to any suitable division of percentile bands. Over each period of analysis (e.g. AP or NP), the BKS can be examined as a frequency histogram of the values for BKS in the manner shown in FIGS. 7A, 7B, 8A, 8B, 9A, 9B, 10A and 10B. The present embodiment recognises that the BKS can be grouped into two super categories referred to herein as a Mobile category and an Immobile category, and that each in turn can be further divided into two subcategories, referred to herein as Active Mobile, Inactive Mobile, Moderate Immobile and Very Immobile, as shown in FIG. 7A. See FIG. 11A for the BKS distribution and Table 3 above for the values observed for BKS in each group.

[0071] In more detail, FIG. 7A is a histogram of BKS units in a Control (non PD) subject from the Active Period (AP). FIG. 7B is a histogram of BKS units in the same subject from the Night Period (NP). In FIGS. 7A and 7B, the x axis is the value of the BKS unit and the Y axis is the number of BKS units with that value. Each histogram shows the four types of BKS categories: Active (0<BKS<44), Inactive (44<BKX<80) and Immobile (80<BKS), which is divided into a Moderate Immobile category (80<BKS<110) and a Very Immobile category (110<BKS).

[0072] The Active.sub.50 value is defined in this embodiment as being the median (and mode) of the Active BKS during the AP. The distribution of the BKS is shown in red in both histograms. It is noted that the distribution of Active BKS in the night period histogram of FIG. 7B is similar to the day period histogram of FIG. 7A. The median BKS of 20.4 for the subject of FIG. 7 is similar to the Active.sub.50 value of 19.6, as is generally the case for normal subjects.

[0073] In FIG. 7A it is notable that during the AP, Inactive BKS are uncommon, as are Moderate Immobile BKS and very Immobile BKS values. In contrast, in FIG. 7B it is notable that during the NP there is a marked peak of Very Immobile BKS values, with the histogram peak occurring at a BKS value of 125. In FIG. 7B the 25.sup.th percentile of BKS values in the Very Immobile range (referred to as the Immobile.sub.25 value) has a value of 114.

[0074] FIG. 7C is a raster plot of 6 six consecutive days denoted AP1 to AP6, showing data from the AP of each day. Each BKS value is shown as a light blue dot in the top row if the BKS is in the Active range (0-44), or as a dark blue dot in the second row if the BKS is in the Inactive range (44-80), or as a black dot in the third row if the BKS is in the Immobile range (BKS>80). A red dot is shown in the fourth row of each raster trace if at least four of the surrounding consecutive BKS values are >80. Each red dot thus indicates that the surrounding 7 consecutive BKS scores reflect the existence of a sleep epoch. It is notable in FIG. 7C that this subject was awake (ie not immobile) and active (most dots light blue) for most of the AP on each of the 6 days observed.

[0075] FIG. 7D is a raster plot of six consecutive evenings, showing data from 22:00-07:00 but with NP shaded in light grey. Each BKS is coloured and positioned in one of four rows, using the same convention described above in relation to FIG. 7C. It is notable in FIG. 7D that the BKS data indicate that this subject was active (with blue dots in the top row) until about 01:00 during night periods NP2-NP5, and was asleep until at least 07:00 during night periods NP1, NP2, NP4 and NP5. On NP6 this subject went to sleep about 3 hours earlier than the other nights and rose shortly after 03:00, which for example might be indicative of a shift worker.

[0076] FIG. 7E provides an enlarged view of a portion of the raster plot of FIG. 7D, illustrating the top row 702 of Active BKS values, the second row 704 of Inactive BKS values, the third row 706 of Immobile BKS values, and the fourth row 708 of sleep epoch data points.

[0077] FIG. 8 shows data obtained from another normal control subject, using the same plotting conventions as FIG. 7. In particular, FIG. 8A shows a histogram of BKS values from the second control subject during the AP (09:00-18:00), and FIG. 8B shows the BKS values from the NP (23:00-06:00). FIG. 8C shows that this person falls asleep most nights around 23:00 and awakes around 06:00 each morning and exhibits relatively normal sleep between those times.

[0078] FIG. 9 shows BKS data obtained from a Person with Parkinson's (PwP). FIG. 9A shows that, during the AP, this person exhibits increased Immobile BKS measures. Moreover, FIG. 9B shows that during the NP a markedly abnormal sleep pattern exists. This is revealed by very little BKS in either the very immobile or immobile range as compared to the controls of FIGS. 7B and 8B. The abnormal sleep is also evidenced in FIG. 9B by way of the increased Inactive and active data throughout the record, as compared to the control subjects of FIGS. 7B & 8B. It is noted that the Active.sub.50 is only modestly elevated in the PwP in FIGS. 9A & 9B, as compared to the Active.sub.50 in FIGS. 7 and 8.

[0079] FIG. 10 shows the data from another PwP. This subject exhibits a marked preponderance of Mobile Inactive BKS Values during the AP, even though there is little Immobility (i.e., little day time sleep). FIG. 10C shows that the subject is late retiring, typically falling asleep around 01:00-01:30. FIG. 10C further shows that this subject has reasonably long periods of sleep as shown by Sleep epochs in the fourth row of each raster plot. However, when FIG. 10B is considered it can be determined that the Sleep Quality is poor as shown by a low Immobile.sub.25 (of BKS=101) and relatively few occurrences of Very Immobile BKS values in FIG. 10B as compared for example to the strong Very Immobile peaks in FIGS. 7B and 8B for normal subjects. Moreover, in FIG. 10B there exists relatively increased Inactive BKS at night as compared to FIGS. 7B and 8B for normal subjects, and the Active BKS has a higher Active.sub.50 (=25.7) than normal subjects, suggesting night time bradykinesia.

[0080] As shown in FIGS. 7A & 7B, in which BKS values are returned on a scale of 0-160, for BKS values greater than 80 the subject is defined as being Immobile. BKS>80 are thus a surrogate marker for daytime sleep. When BKS during the NP are examined in healthy subjects (as in FIG. 7B and FIG. 8B), it is apparent that most Immobile BKS values are part of a Gaussian distribution with very high BKS (typically greater than 110) with a long left sided tail. It is also apparent that People with PD (PwP) are less likely to have this peak of high BKS (see FIGS. 9B and 10B) and have generally lower BKS. We have calculated 25.sup.th percentile of all Immobile BKS values in the NP for each patient and the median of all these values from the 155 Control subjects was produced and called Immobile.sub.25 which was at BKS=111.

[0081] The Moderately Immobile range (MI) is when BKS is between 80-110. The Very Immobile (VI) range is when BKS=111 or greater. Good sleepers have a high proportion of Immobile BKS>110. See Discussion of Sleep Quality Below.

[0082] Day Time Immobility (PTI) is defined as the percentage of time during the AP with Immobility, and has been correlated with polysomnographic recordings of sleep in the daytime. Immobility during the AP is mainly in the MI range when present in normal subjects (FIGS. 7A, 7C and 8A) and in many patients (FIGS. 9A & 10A). Table 3 sets out the normal ranges for PTI as determined from the 155 control subjects and the 72 PD subjects.

[0083] BKS<80 are broadly defined as Mobile. Examination of the Mobile BKS (eg FIGS. 7A, 7B, 8A and 8B) suggests that there are two distributions within BKS<80: a Gaussian distribution typically less than 40-50 BKS and a separate distribution between 40 and 80. Principle Component Analyses (PCA) supported the conclusion that there were indeed two components with BKS<80. FIG. 10A shows an extreme example of a subject clearly exhibiting the separate distribution of BKS in the 40-80 range, independently of and in addition to the Gaussian distribution of BKS<40. Accordingly, this is reflected by the division of Mobile BKS into Active Mobile and Inactive Mobile as shown in FIG. 7A.

[0084] Active BKS are thus BKS measures which fall in the lower Gaussian Distribution. To quantify Active BKS it was therefore first necessary to extract this distribution from the broader data set. To do so, in this embodiment it was assumed that the slope or curve of the BKS values from BKS=0 to the peak (the mode of the distribution outlined by the red line in FIGS. 7A, 7B, 8A, 8B, 9A, 9B, 10A and 10B) represents the left half of the Gaussian distribution sought. This left half is then smoothed from BKS=0 to the mode BKS (shown as the red line in the three graphs in FIG. 7) using an 11-point boxcar filter. This smoothed line is then reflected around the mode to produce the full Gaussian distributed component of the graph (and is shown as the red bell shaped curve in FIGS. 7A, 7B, 8A, 8B, 9A, 9B, 10A and 10B). It is to be appreciated that any suitable method of extracting the lower distribution may be used in accordance with other embodiments of the present invention. All BKS values enclosed by this curve represent Active BKS and the 50.sup.th percentile (and mode) of these values is referred to as the Actives. Without intending to be limited by theory, it is proposed that this subset of the BKS data represents the BKS that are related to and arise from the subject's attempts at being active. The proportion of BKS within the Active Distribution during the AP is referred to as the Percent Time Active (PTA). The upper limit of the Active Distribution is typically about twice the Active.sub.50 and is also the boundary between Active scores and those that lie between this limit and BKS=80 (Inactive Scores). This boundary between Active and Inactive is referred to herein as Boundary.sub.A-I. It is to be appreciated that any suitable value may be selected or determined for Boundary.sub.A-I. FIG. 11A shows plots of the distribution of median BKS (BKS.sub.50), Active.sub.50, and the boundary between Active and Inactive BKS (A-I Boundary) in normal subjects (C) aged greater than 60 and PwP (PD).

[0085] FIG. 11B shows plots of the distribution of median DKS (DKS.sub.50) in normal subjects (C) aged greater than 60 and PwP (PD).

[0086] FIG. 11C is a plot of the difference between median BKS (BKS.sub.50) on the X axis and Active.sub.50 on the Y axis showing these values for both Controls (black dots) and PwP (red triangles). In most cases there is a modest reduction in the Active.sub.50 but on occasions the reduction is large with higher BKS (eg as shown in FIG. 10).

[0087] Inactive. BKS values that lie between the Boundary.sub.A-I and BKS=80. It is assumed that these BKS indicate movement associated with sedentary behaviour and in particular somnolence. They are temporally more common at times when Immobility scores are present and also in the RP when TV and drowsiness often occur. FIG. 10A shows a marked excess of BKS in the Inactivity range. The proportion of BKS within the Inactive Distribution during the AP is referred to as the Percent Time Inactive (PTIn) with control and patient values set out in table 3.

[0088] All BKS categories (Mobile (Active, Inactive) and Immobile (MI and VI) are used in all periods including AP, RP and NP and their percentage time in these categories varies according to which period is being examined (see Table 3 and FIG. 11). FIGS. 11D, 11E, 11F and 11G show the PTA, PTIn and PTI in the AP and NP. As expected PTA is higher in the AP whereas the PTI is higher in the NP. PTI is significantly higher in the day time and lower at night in PwP, compared with controls.

[0089] Night Period and Sleep Scores

[0090] In assessing sleep, we have used the units: BKS>80 or PTI and Sleep Epochs.

[0091] PTI: In the NP is, in effect, the proportion of time in the NP that the subject was immobile. This correlates with sleep in the day but may not be as good a correlation in the NP because people may move (BKS<80) during nocturnal sleep. In more detail, the range of BKS used in this embodiment extends from values of 1 to 150, and there is progressively less energy in the movement as the scores increase. While BKS scores from 80 to 150 do not reflect precisely zero movement, we define herein that the person has moved only if the BKS<80, and that for BKS>80 there exists a range of immobility including both the Immobile and Very Immobile bands.

[0092] Sleep Epoch: To address the issue of movement during sleep, a sleep epoch was produced by taking 7 consecutive BKS values: if the BKS in 4 of the 7 values is >80 then we deem the central epoch as sleep. We then slide the assessment forward in time by 1 BKS epoch and ask again if 4/7 are >80 to score the next BKS as asleep or awake.

[0093] Factors that might be considered in assessing sleep include:

[0094] Efficiency: the extent to which a person slept, throughout the period in which sleep was attempted. This is achieved in the Polysomnography (PSG) lab by measuring time asleep during the period from lights OFF to lights ON. This is difficult at home or otherwise out of the clinical setting with the body worn device of the present invention, because we can only assess when sleep began and not the period over which sleep was attempted (ie in bed and trying to sleep). The choice of the NP being from 23:00 to 06:00 is made because 75% of subjects were asleep within 30 mins of 23:00 and >90% slept till 06:00, as shown in FIG. 11H. In particular, FIG. 11H shows the time that control subjects and PwP retired relative to 23:00 or awoke relative to 06:00. Note that sleep onset before 23:00 or after 06:00 was not assessed. Total as % (right Y axis) refers to the time between first sleep (measured by a train of consecutive sleep epochs: either already asleep at 23:00 or first appearance after 23:00) and last sleep (either before or ending at 06:00) expressed as a percent of the 420 available minutes. Even though the resulting NP is less than the standard 8 hours, most people are asleep over this period (FIG. 11G) and so we assess the amount of sleep by reference to such a definition of NP by the following estimates.

[0095] In all figures bars show the median and interquartile range. These ranges are tabulated in Table 3.

[0096] PTI: This is the proportion of the NP in which BKS>80. While it broadly correlates inversely with time between Offset and onset of sleep, in control the PTI is 25% lower. The PTI is in effect a measure of sleep efficiency

[0097] Sleep Duration. This is the sum of the number of sleep epochs in the NP (multiplied by two to be expressed in minutes). It correlates with PTI with an r.sup.2=0.81.

[0098] PTIn: Subjects who have made movements in their sleep or are awake but attempting sleep, may have BKS<80 and in the Inactive range for that subject.

[0099] Fragmentation: If a person is immobile from sleep for a periodsay 20 minutesthen there will be 10 consecutives 2 minute Sleep Epochs. Such a stretch of consecutive Sleep Epochs is termed a sleep fragment. We postulate subjects who have frequent micro-arousals and periodic limb movements (PLM) are likely to have shorter fragments. In most subjects, the distribution of fragment length is markedly hyperbolic and even though there is a high proportion of short fragments, most of the sleep (immobility) resulting from a small number of long fragments. Thus a measure of fragmentation would be to estimate the proportion of sleep (immobility) resulting from fragments greater than a certain length. To estimate this, we measured the median fragment length (MFL). Control and PD subject values are shown in FIG. 12A.

[0100] Architecture: Sleep Studies suggest that the full sleep architecture requires longer sleep segments and that micro-arousals and periodic limb movements are less frequent during deeper stages of sleep. This suggest that the presence of a proportion of sleep with less movement may reflect better quality sleep architecture. The Immobile.sub.25 (as defined in the preceding) for each Control subject was found, and the median Immobile.sub.25 of all subjects was then calculated (BKS=111). This was used as the boundary between MI (moderate immobility) and VI (very immobile). We then estimated the proportion of time in the NP, that each subject was Very Immobile (VI) and called this Sleep Quality.

[0101] Time Awake. This is related to a number of factors. This includes those related to poor sleep hygiene (late to bed, early rising): factors related to sleep disruption (pain, bladder control etc.): factors related to mood or disrupted sleep regulation (e.g. early awakening from depression). The premise here is that frank awakening will be captured in part by Active BKS (PTA, as described above) rather than PTIn. Arguably Time Awake will be inversely related to Sleep Efficiency.

[0102] Sleep Score.

[0103] Six variables have been described above (PTA, PTI, PTIn, Sleep Duration, MFL, Sleep Quality) but there may be a degree of overlap between them as descriptors of good sleep. Furthermore, each type of sleep disorder is likely to manifest in its own way in such kinetic observations and so a different set of variables might be required to accurately assess different sleep disorders. Thus, the present invention recognises that combining a set of these variables with variable weightings into a single score might better describe disordered sleep, and moreover that different variable weightings can be used to assess different disorders. We use the following steps.

[0104] Step 1. For a Particular Individual, Give Each Variable a Score Ranging from 0-5.

[0105] This is because each variable has a different range (some percentages (0-100) and others in minutes and less than 30 units) and distribution, so they must be normalised if they are to be summed. To achieve this the 10.sup.th, 25th, 50.sup.th, 75.sup.th and 90.sup.th percentile of each variable were found and these were used as a scoring system. A score from 0-5 was given according to Table 4. Note Table 4 provides two inverse options for this conversion, depending on whether the assessment should return higher scores to indicate better sleep, or lower scores to indicate better sleep.

TABLE-US-00004 TABLE 4 HOW SLEEP VARIABLES ARE TRANSFERRRED TO A COMMON SCORING SYSTEM Percentile range High score = good sleep Low score = good sleep 0-10 0 5 10-25 1 4 25-50 2 3 50-75 3 2 75-90 4 1 90-100 5 0

[0106] Step 2. Sum and Weight Each Normalised Variable.

[0107] A Sleep Score for a particular condition (eg PD) could be produced according to the following formula:

Sleep Score(PD)=aPTA+bPTI+c,PTIn+dSleep Duration+eMFL+fSleep Quality

where a, b, c, d, e and fare weightings that might range from 0 (no weight) to some value greater than 1 (to increase the weight). These weights might be determined by inspection, by trial and error or by using machine learning.

[0108] Step 3. Determine the Weightings for a Particular Condition.

[0109] An assumption here is that there already exists a gold standard measure of disordered sleep for each condition. PSG is widely held as the Gold standard for sleep but (a) it is commonly reported subjectively (normal/abnormal); (b) it requires admission to a laboratory and so sleep is in unaccustomed settings with imposed sleep regimen; (c) it has scores for periodic limb movements and arousals but is weighted toward sleep apnoea. A common alternative is to use validated patient reported sleep scales. The Epworth sleepiness score (ESS) is an example for day time sleepiness and the Parkinson's Disease Sleep scale2 (PDSS 2) is an example for sleep in PD. The PDSS 2 is a comprehensive questionnaire that asks about night time sleep patterns and day time sleep patterns. It has the short coming that it is self reported, it covers more than night time sleep and it is non linear. This is important because normal sleep receives a score of 0 even though normal sleep has a wide range of variability and the transition from normal to moderate is by an increment of 2 and so also is the transition from moderate to severe (ie not linear).

[0110] To examine the weightings to apply to variables we have examined the six above-described PKG variables in a) PwP and b) subjects undergoing PSG for a sleep disorder (usually sleep apnoea). We have compared all six variables and, by inspection, chosen to weight those variables that have the greatest variation from controls. We have then iteratively applied different weightings to obtain the greatest correlation with the existing sleep standard (PDSS 2 or PSG).

[0111] Comparison of Sleep Data from Normal Controls and PwP.

[0112] Time of arising and retiring. There was a statistically significant likelihood of PwP to go to bed close to 23:00 and arise before 06:00 but the effect size (ie number of minutes difference) was not very meaningful and this was borne about by a non-significant trend p=0.51) for PwP to have a shorter time attempting sleep (FIG. 11H). Interestingly, there was non-significant trend to difference in the percent time Active (p=0.52 Mann Whitney) due to much greater variability in patients.

[0113] Sleep Efficiency. Differences in the three measures of efficiency (PTI, PTIn and Sleep Duration) are shown in FIGS. 12A and B and in Table 4. In particular, FIGS. 12 A, B, & C show the values for Controls (C, green dots) and PwP (PD, red dots) for all of the variables used to create a sleep score (SS, in FIG. 12 C, right Y axis). These variables are described in the text and are Sleep Duration (A), Median Fragment Length (MFL, FIG. 12B), PTI, PTIn, Sleep Quality (FIG. 12B). PDSS-2 was obtained by questionnaire for most PwP and Controls and is shown in FIG. 12C. In all graphs there was a significant difference between data from Controls and PwP (>0.0001, Mann Whitney). The differences between Controls and PwP were statistically different and with meaningful effect size. In Summary, PwP spent more time awake but Inactive, less time Immobile and with Shorter Sleep Duration.

TABLE-US-00005 TABLE 4 SCORES THAT CONTRIBUTE TO THE SLEEP SCORE PERCENTILE Min 10% 25% Median 75% 90% Max PTA C 5.3 10.2 13.2 17.1 21.1 26.4 63.1 PD 2.3 7.4 9.5 12.6 17.5 24 40.6 PTIn C 5 7 10 13 18 24 41 PD 3.2 8 13 20 28 40 58 PTI C 23 56 61.2 69 75 79 88 PD 1.5 31 45 59.1 75.3 80.3 90.3 Sleep Quality C 31 57 68 77 83 88 93 PD 13 33 45 60 73 81 98 Sleep Duration C 90 217 267 310 350 376 405 PD 2 105 167 257 307 369 405 MFL C 8 19.3 265 38 60 86 227 PD 8 12 16 25 36 82 368 Immobile 25 C 89 100 106 111 115 119 129 PD 85 89 94 101 109 113 142 Sleep Score C 0 5 8 13 17 21 25 revised PD PD 0 0 1 6.5 12 17 25

[0114] Fragmentation was assessed by the Median Fragment Length (MFL) of each Sleep Fragment. This was significantly shorter in PwP (FIGS. 12A and B and in Table 4).

[0115] Sleep Architecture was measured by Sleep Quality, which measures the proportion of Immobile BKS (>80) that are very Immobile (>110, or higher than Immobile.sub.25) (FIGS. 12A and B and in Table 4). This was statistically and meaningfully less in PwP than in Controls.

[0116] Time Awake was measured using PTA (FIG. 12 D). In more detail FIGS. 12 D & E show the PDSS-2 (D) and SS (E) of PwP plotted against duration of disease (years). Controls are shown (C) as a series of bars showing 10.sup.th, 25.sup.th, 50.sup.th, 75.sup.th & 90.sup.th percentiles. The grey bars show the regions above the 90.sup.th percentile (PDSS-2) or below the 10.sup.th percentile (SS). There is a trend toward abnormal scores within the first three years of disease. The mean was not statistically different from Controls although the spread was substantially greater in PD subjects as manifest by the larger interquartile range.

[0117] Comparison of each variable with PDSS 2. PDSS 2 is a recognised Sleep Scale. While it is not expected that there will be very high correlation between each variable and the PDSS 2 they should each have a relevant trend if they are likely to influence a Weighted Sleep Score. Each Variable was compared with the PDSS 2 (FIG. 13A-F, in which circular data points are controls and square data points are PwP). In more detail FIG. 13A shows the relationship between Sleep Score and PDSS-2. The relationship is not significant. FIGS. 13 B, C, D, F & G show the relationship between PDSS 2 and each subcomponent of the Sleep Score. For Sleep Duration and Sleep Quality individually (B and C) and together (G) there is a significant relationship measured by the Fisher's exact test and with the boundaries set at the 25th percentile of each variable (including the PDSS 2). Note that sleep duration and quality are broadly related (E). There is positive relationship between PDSS 2 and MFL but not with PTIn. In most instances, the relationship was not linear so the correlation was tested using a Fishers exact test with the 75th percentile of each relevant test used as the boundary (grey boxes). This reveals that Sleep Quality, Sleep Duration and MFL best correlated with the PDSS 2 although there was a weaker relationship with the other variables.

[0118] Sleep Scores. Using the Sleep Score Formula described above three different Weighted Sleep Scores (WSS) were produced according to the weightings in the table below.

WSS=aPTA+bPTI+c,PTIn+dSleep Duration+eMFL+fSleep Quality

TABLE-US-00006 Weighting PTA PTI PTIn PD MFL SQ WSS A 0 1 1 1 1 1 WSS B 1 0.5 0 1.5 0 1.5 WSS C 0 0 0 1 0 1 WSS D 0 1 0 1 0 1 WSS = Weighted Sleep Score

[0119] WSS C was produced because Sleep Quality and Duration both showed a Good relationship with PDSS 2. WSS B was produced because it was developed for testing against PSG.

[0120] The relationship between each WSS and the PDSS 2 is shown in FIG. 13G-I. This shows that WSS C produced the best relationship with PDSS 2.

[0121] It is notable that MLF also had a good relation with PDSS 2. We predict that using Machine Learning or some other iterative method and a larger data base, better correlation with PDSS 2 and a weighted sleep score will be produced.

[0122] The PDSS-2 and WSS were plotted against duration of disease (FIG. 12F-I). All Measures showed a similar trend for Sleep states to worsen as diseased progressed. WSS C was most similar to the PDSS 2.

[0123] Sleep Scores and Polysomnography. This was a study of 36 subjects who were investigate with a sleep study (mostly for sleep apnoea) and found to have 10 subjects with a normal sleep study (7 as part of a spate research study). Their sleep was grades according to the report from the PSG and scored according to the table below. Note that in Mildthere was some extra comment other normal (eg Normal but fragmented sleep) and so they had a separate category but in some cases could have been normal.

TABLE-US-00007 PSG Mild Severe Controls Mild plus mild-mod Mod minus Severe Score 0 1 2 3 4 5 6

[0124] They wore a PKG for the duration of the sleep study and the same sleep variables described above were examined. However, Sleep Duration was now from the time of lights out to lights on in the laboratory and was expressed as Percent time asleep, because this interval varied in time. The relationship between each variable and the PSG score is shown in FIG. 14. In each of these graphs the Normal subjects are shown as red dots. The grey band represents the scores between the 25th and 50.sup.th percentile of controls and received a score of 1. The orange band represents the scores between the 75th and 50.sup.th percentile of controls and received a score of 2. Above the 75.sup.th received a score of 3 and below the 25.sup.th received a score of 0.

[0125] These variables were then summed into a weighted Sleep Score (WSS) to best optimise the capacity to separate Normal subjects from those classed as having an abnormal PSG (FIG. 6H-J). In the three figures, black dots represent subjects classified as having abnormal sleep by both the PSG and the PKG scores. Those in Burgundy are false normal and in orange false abnormal by the PSG relative to the PSG. The subjects marked by a 1 may well be normal (as are indeed all the subjects marked with a PSG score of 1). The subject marked with a 2 was reported as normal but had a very high Periodic limb movement score on the PSG.

[0126] Noting that the best use of the PKG score with reference to the PSG is as a screening tool, it is best to minimise the cases falsely classed as normal and thus WSS B and WSS C provide similar outcomes. These scores were also the best in terms of the PDSS 2 and it is relevant that the 25th percentile cut off used in the PDSS 2 analyses is almost identical to the cut-off in the relevant comparison with the PSG.

[0127] Accordingly, it can be concluded that we can predict sleep using a weighting of six variables, and the weights of the variables can vary (in this current form from 0-1.5). The choice of weighting is variable and is currently chosen by inspection of the graphs and iterative application to achieve an optimal relationship. However a machine learning approach is a more sophisticated application of the same approach but allows on going improvement as data becomes available.

[0128] These studies further reveal that BKS has ranges (at least four). Immobility induced by sleep is more than just still measured by a higher BKS but includes various grades of two or more levels of stillness as measured by a higher BKS. Quality of sleep has a relationship to the extent of stillness measured by a higher BKS. We believe that this is related to the architecture of sleep.

[0129] Fragmentation is a measure of poor sleep. The length of passage of immobility as measured by the number of consecutive BKS that are greater than some specified BKS value (eg 80 or 110) indicates better sleep. The total duration of sleep (using various analyses of BKS to find a total amount of immobility in a specified period of attempted sleep) is a measure of the quality of sleep.

[0130] The amount of time with Active BKS indicates movements during a night period that suggest either that sleep is not being attempted (poor sleep hygiene) or that movements are intruding into and disrupting sleep (eg REM sleep disorder).

[0131] The amount of Inactive BKS indicates movements during a night period that suggest either that sleep is being attempted but not achieved (insomnia) or that movements are intruding into and disrupting sleep (eg micro-arousals and periodic limb movements).

[0132] These aspects of immobility and Mobility during the night period can be assessed with continuous variables (eg Fragmentation by Median Fragment length), Sleep architecture by measuring the 25th percentile of the BKS value during sleep (ie how still was a person), sleep duration etc). Scores can be given according to the values that represent percentiles (eg 10th, 25th, 50th, 75th, 90th) of control subjects to produce a score for each variable. These variables can be weighted, summed and/or combined by any other suitable mathematical function in order to produce a Sleep Score.

[0133] There is a difficulty in validating these score because of the problem of a gold Standard. One gold Standard is the Polysomnogram and another is sleep scales (eg PDSS 2 of PD). Each has their problems. Polysomnogram. Admitted for one night in unfamiliar surroundings and is highly geared toward measuring sleep apnoea and abnormal sleep (ie not normal sleep). Scales of severity are often binary or descriptive and biased toward sleep apnoea. PDSS 2. Is a questionnaire and biased toward PD sleep problems including daytime sleep and pain.

[0134] Nevertheless, we can show that in the case of PD, each subcomponent is significantly different in PD subjects from controls (albeit with overlapbut not all PD have sleep problems and not all Controls do not have sleep problem). Furthermore our scale worsens as disease progresses and there is a correlation with some subcomponents with the PDSS 2. This may suggest which problems the PDSS 2 favours (or are more important in PD). In the case of PSG, we can predict with high (but not perfect) accuracy who will be abnormal. Our conclusion is that the existing measures are disease specific and that we can provide sub-scores and total scores that indicate sleep pathology and can be used qualitatively and quantitatively. Using the weightings and the different measures is novel and of value.

[0135] While the described embodiments are directed to the identification of normal and abnormal sleep in the context of PD in particular, it is to be appreciated that alternative embodiments of the invention may be applied to identify sleep abnormalities arising from other conditions and in particular non-apnoea sleep abnormalities. For example, it is to be noted that the non-PD control group data, such as found in FIGS. 7D and 8C, included instances of sleep disturbances which may in other embodiments of the invention be correlated with other conditions by derivation of suitable sleep variable weightings optimised for such conditions in the manner described herein. In particular, non-apnoea conditions may be assessed in this manner.

[0136] Reference herein to a module may be to a hardware or software structure which is part of a broader structure, and which receives, processes, stores and/or outputs communications or data in an interconnected manner with other system components in order to effect the described functionality.

[0137] Some embodiments of the invention may employ kinetic state or sleep state assessment in accordance with any or all of the teaching of International Patent Publication No. WO 2009/149520 by the present applicant, the content of which is incorporated herein by reference.

[0138] Thus accelerometry using the Parkinson Kinetigraph (PKG, from Global Kinetics) can be used to distinguish between normal and abnormal sleep in Parkinson's Disease (PD)

[0139] It will be appreciated by persons skilled in the art that numerous variations and/or modifications may be made to the invention as shown in the specific embodiments without departing from the spirit or scope of the invention as broadly described. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive.