A Continuous Wavelet Transform and Classification

298

IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 17, NO. 3, JUNE 2009

A Continuous Wavelet Transform and Classification Method for Delirium Motoric Subtyping Alan Godfrey, Member, IEEE, Richard Conway, Member, IEEE, Maeve Leonard, David Meagher, and Gearóid M. Ólaighin, Senior Member, IEEE

Abstract—The usefulness of motor subtypes of delirium is unclear due to inconsistency in subtyping methods and a lack of validation with objective measures of activity. The activity of 40 patients was measured over 24 h with a discrete accelerometer-based activity monitor. The continuous wavelet transform (CWT) with various mother wavelets were applied to accelerometry data from three randomly selected patients with DSM-IV delirium that were readily divided into hyperactive, hypoactive, and mixed motor subtypes. A classification tree used the periods of overall movement as measured by the discrete accelerometer-based monitor as determining factors for which to classify these delirious patients. This data used to create the classification tree were based upon the minimum, maximum, standard deviation, and number of coefficient values, generated over a range of scales by the CWT. The classification tree was subsequently used to define the remaining motoric subtypes. The use of a classification system shows how delirium subtypes can be categorized in relation to overall motoric behavior. The classification system was also implemented to successfully define other patient motoric subtypes. Motor subtypes of delirium defined by observed ward behavior differ in electronically measured activity levels. Index Terms—Activity, classification, continuous wavelet transform, delirium, subtypes.

I. INTRODUCTION

T

HE use of accelerometry as a means by which to monitor the activity profiles of various clinical groups has been well documented with both commercial and discrete accelerometer-based devices having been tried and tested in laboratory, clinical, and home dwelling environments to accurately record such activities as sitting, lying, standing, and stepping [1]–[5]. Studied groups have ranged from those with ailments such as back pain and venous ulceration to chronic heart disease [2], [6], [7]. The findings presented in these studies have successfully shown the use of accelerometry to determine the activity profiles of these groups and their reduced/altered activity compared to healthy controls. Manuscript received October 23, 2008; revised January 08, 2009; accepted February 23, 2009. First published June 02, 2009; current version published July 06, 2009. A. Godfrey is with the School of System Engineering, University of Reading, RG6 6UR Reading, U.K. (e-mail: [email protected]). R. Conway is with the Department of Electronic and Computer Engineering, University of Limerick, Limerick, Ireland. M. Leonard and D. Meagher are with the Department of Psychiatry, MidWestern Regional Hospital, Limerick, Ireland. G. M. Ólaighin is with the Electrical Engineering Department, National University of Ireland, Galway (NUIG), Ireland, and the National Centre for Biomedical Engineering and Science, NUIG, Galway, Ireland. Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TNSRE.2009.2023284

An initial preliminary study presented by Leonard et al. [8] has shown that the use of accelerometry may also be adapted to provide useful information within a particular study group. The group presented by Leonard were consecutive cases of DSM-IV delirium [9]. Delirium is the umbrella term applied in both ICD-10 [10] and DSM-IV to denote acute generalized disturbances of cognitive function and reflects a complex neuropsychiatric syndrome that is highly heterogenous in causation and clinical presentation. Delirium is a complex neuropsychiatric syndrome occurring in 11%–42% of general medical inpatients [11], and up to 50% of hospitalised elderly [12]. It is associated with elevated morbidity and mortality but remains understudied with up to two-thirds of cases missed in clinical practice [13]. Recent work indicates that disturbances of motor function are almost invariably (94%) present during any 24 h of delirium [14] and as such are considered a core feature of the syndrome. Clinically defined subtypes may promote better understanding of delirium by allowing more targeted studies of pathophysiological underpinnings and treatment needs. It was shown within Leonard’s preliminary study that accelerometry may be used as an objective means to determine these motoric subtypes based on quantitative aspects of human movement [8]. Human movement results in a nonstationery signal [15] as not only do its quantitative aspects change over time but so too do its qualitative properties. A suitable method to examine the qualitative components in such a nonstationery signals is the use of the multiresolution analysis. The multiresolution analysis technique adopted for this study was the continuous wavelet transform (CWT). Wavelet analysis provides a variety of different probing functions that can be used in a compressed or enlarged state as well as translations. Wavelets do not exist at a specific time or a specific frequency and as such provide a compromise between time and frequency localization. This is due to the fact that they are localized well in both time and frequency but not precisely localized in either [16]. The application of the CWT in this study will be a useful tool in primarily examining the qualitative aspects of human movement generated per delirium subtype, while also extracting some quantitative components. Where quantitative analysis have successfully been shown for the potential to determine subtype [8], the qualitative components may highlight new aspects of human motion applicable to delirium motoric subtyping. In this paper we describe the use of a discrete accelerometer-based activity monitoring system on patients with varied motoric presentations of delirium and we examine the use of the CWT as a suitable means to determine delirium motoric subtypes based on overall movement by comparing it to the clinical motoric checklists for delirium. (From this we also aim to determine the most suitable wavelet method for successful subtype

1534-4320/$25.00 © 2009 IEEE Authorized licensed use limited to: NATIONAL UNIVERSITY OF IRELAND GALWAY. Downloaded on September 7, 2009 at 10:37 from IEEE Xplore. Restrictions apply.

GODFREY et al.: A CONTINUOUS WAVELET TRANSFORM AND CLASSIFICATION METHOD FOR DELIRIUM MOTORIC SUBTYPING

determination.) A classifier system was also implemented based upon nonlinear regression models to subsequently define patient subtypes based upon the findings of the CWT method.

II. METHOD This study was conducted at Milford Hospice Palliative Care Centre, Limerick, Ireland. Forty consecutive patients receiving palliative care were selected for accelerometry monitoring. Clinical assessments to rate delirium phenomenology and severity, assesses medical morbidity and etiology, and define motor subtypes were performed at the end of the accelerometry monitoring period and encompassed the previous 24 h. All prescribed medications were noted including benzodiazepines, opioids, antipsychotics, stimulants, antidepressants, and steroids. A. Informed Consent The procedures and rationale for the study were explained to all patients but because patients had DSM-IV delirium at entry into the study it was presumed that many were not capable of giving informed written consent. Ethics committee approval was given to augment patient assent with proxy consent from next of kin (where possible) or a responsible caregiver in accordance with the Helsinki Guidelines for Medical research involving human subjects. It was agreed that if at any time during the study if monitoring devices were deemed by carers, family, or nursing staff to be causing distress or discomfort in any way that they would be immediately removed and monitoring discontinued. B. Clinical Assessments The clinical assessment scales used within this study consisted of the following. 1) A New Subtyping Scheme: Motor subtypes were identified according to criteria from a new subtyping scheme derived from key elements of previous methods. In contrast to previous methods of motor subtyping in delirium, this scale emphasizes actual disturbances of motor behavior rather than the associated disturbances (e.g., psychotic symptoms, affective disturbances, aggression) that can occur in delirium that do not directly relate to motor activity levels. It consists of 11 items (four hyperactive and seven hypoactive). Motor subtypes are denoted according to the presence of two or more items for hyperactive and/or hypoactive subtype, respectively. It is simple, brief and can be rated by nurses [14], [17]. 2) Delirium Rating Scale—Revised 98 (DRS-R98): Delirium phenomenology and severity were assessed using the Delirium Rating Scale-Revised-98 (DRS-R98), which was designed for phenomenological assessment using anchored descriptive item ratings on Likert scales for its 16 items [18]. It has high interrater reliability, validity, sensitivity, and specificity for distinguishing delirium from among mixed neuropsychiatric populations including dementia, depression, and schizophrenia [18].

299

DRS-R98 severity scale scores range from 0 to 39 with higher scores indicating more severe delirium and a cutoff score consistent with a diagnosis of delirium. The items of particular interest in this study are the DRS-R98 motoric items #7 and #8. These items rate levels of motor agitation and retardation, respectively, and can allow motor subtyping with a score of two or more on either indicating motor subtype criteria. 3) Memorial Delirium Assessment Scale (MDAS): The MDAS is a 10-item clinician-rated scale with a range of 0–30 designed to diagnose and rate the severity of delirium in medically ill patients, including those with cancer [19]. Item #9 denotes motor profile and defines motor subtyping. 4) Cognitive Test for Delirium (CTD): Cognitive function was assessed with the CTD [20] which includes five neuropsychological domains—orientation, attention, memory, comprehension, and vigilance. Scores range between 0–30 with higher scores indicating better cognitive function. The CTD reliably differentiates delirium from other neuropsychiatric conditions including dementia, schizophrenia, and depression [21]. 5) Delirium Etiology Checklist (DEC): The Delirium Etiology Checklist [22] is a structured tool that uses all sources of clinical information to document a wide variety of medical etiologies and then applies a weighted approach for each category of potential causes of the delirium. Each etiological category is rated as ruled out (0), present but apparently not contributory (1), present and possibly contributory (2), likely cause (3), and definite cause (4) according to clinical judgment of the primary physician based on clinical history and investigation. The DEC allows for multiple concomitant causes as contributing etiologies for delirium. 6) Ease of Ward Management Scale (EOWM): The EOWM is a four point checklist that outlines the ease with which patients no problems; can be taken care of by nurses on the unit ( mild problems manageable with minimal intervention/ medication; moderate problems necessitating more than severe problems requiring sedation minimal medication; or special nursing care).

C. Discrete Accelerometer-Based Monitoring Setup Each patient wore a discrete accelerometer-based activity monitor continuously for a 24-h period. The discrete activity monitor consisted of a dual-axis accelerometer1 sensor that was positioned on the lateral aspect of the mid-thigh, which measured vertical ( ) and transverse ( ) accelerations. The sensor was connected to a lightweight (86-g) data-logging device2 that recorded the raw acceleration data at a sampling frequency of 50-Hz and at a resolution of 12 bits. The data logger was positioned on the anterior of the thigh and both the sensor and data logger were held in place by means of a disposable cuff3 underneath the subjects clothes (Fig. 1). The raw accelerometer data were saved to the data logger’s internal 1ADXL322, 2Biomedical

Analog Devices BV, Limerick, Ireland. Monitoring Ltd., Wolfson Centre, Glasgow, U.K.

3Novaform Conforming Bandage, Midland Bandages Limited, Tullamore, Offaly, Ireland.

Authorized licensed use limited to: NATIONAL UNIVERSITY OF IRELAND GALWAY. Downloaded on September 7, 2009 at 10:37 from IEEE Xplore. Restrictions apply.

300


TABLE I SELECTION OF WAVELETS USED FOR DIRECT COMPARISON WITH THE ACCELEROMETER SIGNALS. EACH WAVELET WAS APPLIED TO DUAL-AXIS ACCELEROMETER DATA TO DETERMINE THE MOST SUITABLE FOR PATIENT MOTORIC SUBTYPING

Fig. 1. Attachment of the discrete accelerometer-based system.

removable memory card and downloaded to a computer for analysis after completion of recording. D. Data Analysis As human movement results in a nonstationery signal [15] it was suggested that a suitable method to examine the qualitative components in such a nonstationery signals would be the use of multiresolution analysis technique. One such multiresolution analysis technique is the use of the continuous wavelet transform (CWT). E. Continuous Wavelet Transform The CWT of a 1-D signal provides a decomposition of the signal at different scales (inversely related to frequencies) and is defined by (1). The transformed signal is a function of two variables, and , the translation and the scale, respectively, where defines the probing function (wavelet) (1) The resulting wavelet coefficients generated from this equation describe the correlation between the waveform and the wavelet used at the various translations and scales. Alternatively it may be stated that the coefficients provide the amplitudes of a series of wavelets, over a range of scales and translations, that would need to be added together to reconstruct the original signal [16]. The Matlab Wavelet Toolbox was used to calculate the different wavelet transforms used in this study4. It was hypothesized that the correct subtyping of an initial group of patients by the CWT method to determine the coefficients generated per readily defined subtypes and subsequent use of a classification tree method would allow for the remaining cohort to be correctly subtyped. Therefore, data analysis was performed with two studies as follows. 1) Study 1: Initial Application of the CWT and Correct Mother Wavelet Determination: Before the CWT analysis could progress, it was necessary to determine the most suitable mother wavelet that would give the best patient subtyping. Three patients that readily met motoric subtype criteria (defined as being hyperactive, hypoactive and mixed by the motoric items on the DRS-R98, MDAS and third subtyping scheme) were chosen at random and labeled Group 1. Study 1 applied 4MathWorks

Inc., Natick, MA

the CWT with numerous mother wavelets to the biaxial accelerometer data collected from this group. The maximum coefficients generated over a range of scales per mother wavelet were plotted. The plot yielding the most distinct separation between patients resulted in the most suitable mother wavelet being chosen to continue analysis for the remaining cohort. With the most suitable mother wavelet chosen, the CWT was reapplied to Group 1. A high pass step-filter (with a predetermined cutoff value) was applied to the coefficient values to reduce the number of coefficients generated for ease of computation. The number of remaining coefficients per scale along with the minimum, maximum, and standard deviation of these values were calculated. These four values were termed as motoric classification values (MCV’s). This procedure provided these four values per CWT scale for which to create a classification tree to subtype the remaining cohort. F. Classification Tree Generation An automatically generated classification tree (decision tree) was generated using the Matlab Statistics Toolbox4 with the function “treefit.” The rule for splitting was based upon one of the standard options of Gini’s index [23]. This method progressively looks for the largest class in the data set and tries to isolate it from the rest of the data [24]. The MCV’s from Group 1 were used to build and train a classification tree for delirium subtyping. Initially the full range of CWT scales returning a different subtype per scale was computed. Subsequent investigation determined the most suitable range of scales from which to choose correct patient subtype. 1) Study 2: Application of the Classification Tree: Once the classification tree was created the MCV’s from the remaining cohort of patients were entered to establish the patients motoric subtype. The complete program control flow is as follows. Study 1 1) The recorded accelerometer signals are input from an ASCII representation of the recorded data. 2) The program calibrates the signal by calculating and removing any offset in the recorded signal (Section II-G). 3) A 1-s window is moved over the signal (1-s segmentation) and the mean corresponding to each window is calculated. This generates an acceleration value per second to speed up overall computation of the CWT.



301

TABLE III DEMOGRAPHIC AND CLINICAL DATA FOR GROUP 1. SHOWN ARE THE TOTAL NUMBER OF MEDICATIONS, NUMBER OF PSYCHOTROPIC MEDICATIONS, DEC SCORE, DRS-R98 SEVERITY SCORE, CTD SCORE, AND EOWM RATING FOR PATIENTS A, B, AND C – HYPERACTIVE, HYPOACTIVE, AND MIXED, RESPECTIVELY

Fig. 3. An example of a period (1 h) of raw uni-axial acceleration signals (a –vertical acceleration) for Patient A–Hyperactive, Patient B–Hypoactive, Patient C–Mixed.

Fig. 2. Flow diagram of the Matlab program.

TABLE II MOTOR SUBTYPE CRITERIA RATINGS FOR GROUP 1 DURING THE 24-H OF MOTION ANALYSIS. PRESENTED ARE PATIENT A, B, AND C WHO ARE READILY DEFINED MOTORIC SUBTYPES ACCORDING TO THE MOTORIC ASPECTS OF THEIR SCALES I.E., DRS-R98 #7 AND #8, MDAS #9 AND THIRD SCHEME

4) The CWT was applied from scales 1–200 (at successive intervals of 2).

5) The CWT was repeated for numerous wavelets (Table I) to find the optimum wavelet method for delirium patient subclassification. 6) The maximum coefficient generated per scale were chosen and plotted (Fig. 5–Section III-A). These coefficients represent the correlation between the waveform and the wavelet at various translations and scales [16]. 7) The resulting maximum coefficients plots per wavelet were analyzed by visual inspection and the most suitable plot over the range of scales was chosen as the most suitable wavelet for subtype determination. 8) The CWT with the most suitable wavelet was then reapplied to Group 1. The resulting coefficients were high-pass step filtered using a predetermined cutoff value. 9) The remaining number of coefficients, minimum value, maximum value and standard deviation of values per CWT scale were entered into a classification system and matched to the corresponding subtype.


302


Fig. 4. (A) 24-h CWT plots (db1 wavelet) for Group 1 patients: (a) Hyperactive–Patient A, (b) Hypoactive–Patient B, and (c) Mixed–Patient C. (B) Hourly CWT plots (db1 wavelet) for Group 1 patients: (a) Hyperactive–Patient A, (b) Hypoactive–Patient B, and (c) Mixed–Patient C. The hourly representations are equivalently match hours for each patient as a direct comparison between subtypes. The raw accelerometer signals are shown in Fig. 4.

Study 2 1) The classification system subsequently determined other patient subtypes. Controls were also classified to highlight compatibility in activity levels. The program flow is summarized in Fig. 2.

G. Calibration The sensors were calibrated prior to patient attachment by measuring the accelerometer signal under controlled conditions. This was achieved by rotating the sensor to provide a signal , 0 g, and . The Matlab prooutput corresponding to gram then corrects any offset in the raw acceleration data from the patient.



303

H. Nondelirious Control Comparison Once delirium subtyping was complete a nondelirious control group were assessed using the exact same methodologies. Their data were analyzed by the CWT (db1) and subsequently by the classification tree. The outcome from this was to highlight which subtype they most resembled for motoric activity. III. RESULTS Forty patients were selected for accelerometry. Of these, complete 24-h readings were available for 34 patients and analyses were confined to this group. The reasons for nonavailability of accelerometry data were premature termination of monitoring, concerns from nursing staff or family that the device was potentially causing patient distress and necessity of medical procedures that required device removal. Of the 34 years , 9 female patients included, 25 (16 male, years ) met criteria for DSM-IV delirium while nine (6 male, years , 3 female years ) were nondelirious comparison subjects with equivalent medical diagnosis receiving treatment in the same setting. A. Study 1: Application of the CWT and Correct Mother Wavelet Selection Table II shows motor behavior according to DRS-R98 (items #7 and #8), MDAS (item #9), and third motor subtyping scheme for the three patients selected for Group 1. From these three assessment scales patients from Group 1 were classified as hyperactive (Patient A), hypoactive (Patient B), and mixed (Patient C). Table III depicts demographic and clinical data for these three patients. The hypoactive patient was older and receiving more medications compared to the hyperactive and mixed subtypes but there were no significant differences between the subtypes in relation psychotropic agent exposure. The CWT was then applied to these patient’s raw accelerometer data (Fig. 3) with numerous wavelets tested (Table I). The CWT was applied over the full recording period of 24-h for these patients for the multiple wavelets from scales 1–200 (intervals of 2). The 3 CWT (db1) images to the left of Fig. 4 represent the full 24-h monitoring period for each of the three patients from Group 1. On initial inspection there appears to be little difference between the subtypes of hyperactive and mixed, while the hypoactive subtype is clearly distinguishable. However, when an hourly representation of this data is examined (three CWT images to the right of Fig. 4) there is a clear separation between subtypes. Down sampling as a result of finding the mean value per second was still found to give a sufficient quantity of data (every second over 24 h) to analysis to determine patient subtype for this group. The maximum coefficient value per scale from the entire coefficients generated per wavelet were selected and plotted. The resulting plots were examined and the plot that gave greatest distinction and clarity of separation between subtypes across the entire range of scales was chosen as the wavelet most suitable for subtyping. A number of the wavelets tested were successful in defining each subtype to some degree. However, the wavelet that gave the greatest separation between subtypes

Fig. 5. Maximum coefficient plots for wavelet db1 for Patient A, Patient B, and Patient C.

was the Daubechies 1 (db1) wavelet. The maximum coefficient values from scales 1–200 for the initial assessment Group 1 are shown in Fig. 5. Thus, the initial three selected patients were clearly distinguished using the CWT method with db1 wavelet and these findings concur with the motor rating items from both the DRS-R98 and MDAS assessments along with the observed ratings by nurses from the third subtyping scheme. 1) Classification Tree Outcome: With the appropriate CWT method determined, this was again reapplied to Group 1 to construct a classification tree. The CWT coefficients from these patients were then high-pass step filtered and the appropriate MCV’s over the range of scales were calculated. A classification tree was created (Statistics Toolbox) with these input variables and the corresponding subtypes of Group 1. The portion correctly classified for Group 1 was equal to 98.7%. Fig. 6 shows the resultant classification tree created from Group 1. B. Study 2: Application of the Classification Tree Once the MCV input variables were determined for each of the remaining patients these values were then passed to the classification tree to determine their corresponding subtype. As each scale generated an individual subtype a selection were available per patient over the entire scale range. Upon inspection of these results it was found that scales 3–31 gave the optimum return from which to choose the patient subtype, with higher scales offering little to separate patients. This range corresponded to 15 different scale (frequency) components from which to choose the correct subtype. As a result the classification tree not only offers a distinction between subtypes but also highlights the potential for severity of motor subtype and evidence of variation between the three. Table IV shows the clinical assessment (DRS-R98, MDAS, Third Scheme) outcomes for each patient and the resultant CWT and classification tree outcomes as determined from this scale range. The dominant classification tree output was taken as the patient subtype.


304


Fig. 6. Classification tree for delirium subtyping based upon CWT (db1). Key: Size-Sag = Number of coefficients, sagittal plane. Std-Tran = Standard deviation of coefficients, transverse plane. Min-Tran = Minimum value of coefficient, transverse plane. Size-Tran = Number of coefficients, transverse plane, Std-Sag = Standard deviation of coefficients, sagittal plane. Max-Sag = Maximum value of coefficient, sagittal plane.

The CWT and classification tree predicted subtypes have a 96% accuracy when compared to any of the outcomes from the three assessment scales. Individual scale comparison shows a 48% agreement between the DRS-R98 and Third scheme to the classification tree, while the MDAS shows a 52%–64% agreement. Table V depicts demographic and clinical data for the entire cohort and control patients. The hypoactive patients were compared to the hyperactive and mixed subtypes. There were no significant differences between the subtypes in relation to overall medication exposure or psychotropic agent. (Delirium subtypes were receiving slightly higher doses when compared to the nondelirious controls). EOWM was higher for the hyperactive subtypes, typical for this category. Table VI shows the resulting outcomes for the control subjects when the same methodologies applied. All but one control patient displays activity similar to that of the mixed subtype with underlying degrees of hyperactivity. IV. DISCUSSION AND CONCLUSION A discrete accelerometer-based monitor was applied to 40 consecutive patients receiving palliative care. Monitoring of these patients lasted for 24-h for direct comparison with 24-h based clinical assessments. Twenty-five of these patients met criteria for DSM-IV delirium while nine were nondelirious comparison subjects with equivalent medical diagnosis. Patients with delirium generally tolerated accelerometry without any major adverse effects or significant difficulty even though some

patients had periods of moderate to severe motor activity/agitation. The location and small physical size of the devices, which were located under clothing, limited the patient’s awareness of their presence and reduced the tendency to disturb or interfere with them. As human movement results in a nonstationery signal it was suggested that the multiresolution analysis was the most suitable method for signal analysis due to its ability to overcome resolution problems encountered by other Fourir techniques for such nonstationery signals [15]. The multiresolution technique used in this study was the continuous wavelet transform (CWT) and was applied to the 25 delirium patients eligible for analysis as they completed the required full monitoring period of 24-h for direct comparison with clinical assessment scales. The Delirium Rating Scale Revised-98 (DRS-R98) items number 7 and 8, Memorial Delirium Assessment Scale (MDAS) item number 9, and a third subtyping scheme were used to determine delirium subtype clinically at the end of accelerometry monitoring. An initial study was undertaken with three patients (Group 1) that readily met subtype criteria according to clinical assessment. Study one concluded that the continuous wavelet transform was a suitable method to determine delirium subtypes with daubechies 1 (db1) wavelet the most suitable mother wavelet for this purpose. Study two evaluated the use of a classification tree for the process of determining subsequent patient subtype based on the findings of the CWT from Group 1. Initial testing of this classification tree returned a success rate of 98.7% for correct classification within Group 1.



305

TABLE IV PATIENT SUBTYPES ACCORDING TO THE THREE CLINICAL ASSESSMENT SCALES AND RESULTING CWT AND CLASSIFIER SYSTEM. CLASSIFIER SYSTEM RATED THE PATIENT OUT OF 15 SCALES (FREQUENCY RANGES) FOR WHICH TO DETERMINE THE MOST ACCURATE MOTORIC SUBTYPE. THIS TECHNIQUE HIGHLIGHTED PERIODS OF ALTERED MOTORIC ACTIVITY, WHICH SATISFY EACH CLINICAL ASSESSMENT SCALE

* The clinician noted if the patient, while diagnosed as mixed, displayed a shift/tendency in subtype towards the hyperactive or hypoactive state). TABLE V DEMOGRAPHIC AND CLINICAL DATA FOR THE REMAINING PATIENTS THAT WERE CLASSIFIED ACCORDING TO THE CWT AND CLASSIFIER SYSTEM. : ) (*

Mean 6 Std Deviation

TABLE VI RESULTING COMPARISON SCHEME FOR THE CONTROL PATIENTS TO DELIRIUM SUBTYPES. DUAL-AXIS ACCELEROMETER DATA WAS OBTAINED FROM A CONTROL GROUP TO HIGHLIGHT THE NORMALISED (NONDELIRIOUS) MOTORIC ACTIVITY THAT IS COMMON IN AN ENVIRONMENT SIMILAR TO THE DELIRIOUS GROUP. THIS MAY HIGHLIGHT HOW THE CONDITION COULD GO UNNOTICED AND ULTIMATELY UNTREATED

With the successful implementation of the CWT with db1 wavelet and classification tree, the same methodologies were applied to the remaining cohort to determine their subtype. The

classification tree output defined a particular delirium subtype per scale from the CWT for each patient. Scales 3–31 (with intervals of two chosen here) were chosen as the best range from which to determine patient subtype in accordance with the clinical assessment scales (DRS-R98, MDAS, and Third Scheme). This gave 15 different frequency ranges of movement from which the subtypes of hyperactive, hypoactive, and mixed were compared. The dominant spread of subtype across these frequency ranges returned an overall successful subtyping of 96% in comparison with the clinical assessment scales. As the classification tree presents the subtype for each patient for 15 scale (frequency) ranges this allows for a more detailed view of the change in variation of general motoric activity in the qualitative aspect over the entire monitoring period of 24 h. When the control group, were assessed with the same procedures the outcome from the classification tree was to define the large majority (8/9) of the control group as having properties similar to the mixed subtype with underlying degrees of hyperactivity. The use of this system highlights the potential use of accelerometer-based monitoring as a suitable and appropriate means for which to determine delirium motoric subtypes. Accelerometer-based devices, both commercial and discrete systems used for research (such as the one used in this study), offer the potential for unbiased and objective measures of motoric activity. A particular advantage of accelerometer-based monitoring for this group is the discrete and continued assessment that it provides. This “remote monitoring” technique has the ability to detect periods of activity/agitation that could affect subtyping which may go unnoticed by the palliative care staff


306


or clinician. This could lead to more accurate patient treatment and to better patient care. The remote monitoring aspect also allows for nursing staff to concentrate their efforts on patients who readily need attention and therefore more efficient time ward management. The dominant subtype returned by the classification tree in this study was chosen here as the resulting subtype for the patient. The use of the continuous wavelet multiresolution analysis has also highlighted the fact that altering intensities or types of movement between subtypes do not differ in any great detail. This qualitative aspect of patient movement may then not readily impact upon patient subtyping and therefore more quantitative components of patient movement should be examined. This study does have some limitations. As the CWT and classification system are trained from the three concurring assessment scales (all three scales agree on subtype) the subsequent subtyping of the other delirium patient’s maybe accepted with a high degree of confidence. However, a lager study group ( ) for which to train the classification tree from which all patients readily meet subtype criteria may further refine the subtyping process using this method. Also, while we did measure medication exposure we did not examine the issue of recent dose alterations, dose equivalents, and the potential for accumulation of agents with longer half lives especially in patients with organ failure. Further work, involving larger numbers of nonpalliative care populations will clarify the extent to which these findings can be generalised to delirious patients in general. ACKNOWLEDGMENT The authors would like to thank the staff and patients at Milford Hospice for their support and Analog Devices BV, Limerick, Ireland, for the supply of accelerometer devices. The authors would also like to thank Dr. F. Hwang and the University of Reading for their support during the completion of this paper. REFERENCES [1] K. Culhane, G. Lyons, D. Hilton, P. Grace, and D. Lyons, “Long-term mobility monitoring of older adults using accelerometers in a clinical environment,” Clin. Rehabil., vol. 18, pp. 335–343, 2004. [2] H. van den Berg-Emons, H. B. J. Bussmann, A. Balk, D. Keijzer-Oster, and H. J. Stam, “Level of activities associated with mobility during everyday life in patients with chronic congestive heart failure as measured with an “Activity monitor”,” Phys. Therapy, vol. 81, pp. 1502–1511, 2001. [3] C. Ni Scanaill, S. Carew, P. Barralon, N. Noury, D. Lyons, and G. M. Lyons, “A review of approaches to mobility telemonitoring of the elderly in their living environment,” Ann. Biomed. Eng., vol. 34, pp. 547–563, 2006. [4] G. Lyons, K. Culhane, D. Hilton, P. Grace, and D. Lyons, “A description of an accelerometer-based mobility monitoring technique,” Med. Eng. Phys., vol. 27, pp. 497–504, 2005. [5] A. Godfrey, K. Culhane, and G. Lyons, “Comparison of the performance of the activPALTM Professional physical activity logger to a discrete accelerometer-based activity monitor,” Med. Eng. Phys., vol. 29, pp. 930–934, 2006. [6] J. Bussmann, Y. van de Laar, M. Neeleman, and H. Stam, “Ambulatory accelerometry to quantify motor behaviour in patients after failed back surgery: A validation study,” Pain, vol. 74, pp. 153–161, 1998.

[7] M. Clarke-Moloney, A. Godfrey, V. O’Connor, H. Meagher, P. Burke, E. Kavanagh, P. Grace, and G. Lyons, “Mobility in patients with venous leg ulceration,” Eur. J. Vascular Endovascular Surg., vol. 33, pp. 488–493, 2007. [8] M. Leonard, A. Godfrey, M. Silberhorn, M. Conroy, S. Donnelly, D. Meagher, and G. ÓLaighin, “Motion analysis in delirium: A novel method of clarifying motoric subtypes,” Neurocase, vol. 13, pp. 272–277, 2007. [9] American Psychiatric Association: Diagnostic and Statistical Manual of Mental Disorders 4 ed. Washington, DC, Amer. Psychiatric Assoc., 2000. [10] World Health Organization, Mental and Behavioural Disorders (F00-F99). The International Classification of Diseases ICD-1010th ed. Geneva, World Health Org., 1992. [11] N. Siddiqi, A. House, and J. Holmes, “Occurrence and outcome of delirium in medical in-patients: A systematic literature review,” Age Ageing, vol. 35, pp. 350–64, 2006. [12] M. Cole, “Delirium in elderly patients,” Amer. J. Geriatric Psychiatry, vol. 12, pp. 7–21, 2004. [13] S. Inouye, “The dilemma of delirium: Clinical and research controversies regarding diagnosis and evaluation of delirium in hospitalized elderly medical patients,” Amer. J. Med., vol. 7, pp. 278–88, 1994. [14] D. Meagher, M. Moran, B. Raju, D. Gibbons, S. Donnelly, J. Saunders, and P. Trzepacz, “Motor symptoms in 100 patients with delirium versus control subjects: Comparison of subtyping methods,” Psychosomatics, vol. 49, pp. 300–308, 2008. [15] B. Najafi, K. Aminian, A. Paraschiv-Ionescu, F. Loew, C. J. Bula, and P. Robert, “Ambulatory system for human motion analysis using a kinematic sensor: Monitoring of daily physical activity in the elderly,” IEEE Trans. Biomed. Eng., vol. 50, no. 6, pp. 711–723, Jun. 2003. [16] J. Semmlow, Biosignal and Biomedical Image Processing: Matlab Based Applications. New York: Marcel Dekker, 2004. [17] D. Meagher, M. Leonard, and P. Trzepacz, “Validation of a new motor subtype scheme for delirium,” World J. Biol. Psychiatry, vol. 8, p. 214, 2007. [18] P. Trzepacz, D. Mittal, R. Torres, K. Kanary, J. Norton, and N. Jimerson, “Validation of the delirium rating scale- revised-98: Comparison with the delirium rating scale and the cognitive test for delirium,” J. Neuropsychiatry Clin. Neurosci., vol. 13, pp. 229–242, 2001. [19] W. Breitbart, B. Rosenfeld, A. Roth, M. Smith, K. Cohen, and S. Passik, “The Memorial Delirium Assessment Scale,” J. Pain Symptom Manage., vol. 13, pp. 128–137, 1997. [20] R. Hart, J. Levenson, and C. Sessler, “Validation of a cognitive test for delirium in medical ICU patients,” Psychosomatics, vol. 37, pp. 533–546, 1996. [21] R. Hart, A. Best, C. Sessler, and J. Levenson, “Abbreviated cognitive test for delirium,” J. Psychosomatics Res., vol. 43, pp. 417–423, 1997. [22] D. Meagher, M. Moran, B. Raju, D. Gibbons, S. Donnelly, J. Saunders, and P. Trzepacz, “Phenomenology of 100 consecutive adult cases of delirium assessed using standardised measures,” Brit. J. Psychiatry, vol. 190, pp. 135–41, 2007. [23] L. Breiman, J. Freidman, R. Olshen, and C. Stone, Classification and Regression Trees. Belmont, CA: Wadsworth Int. Group, 1984. [24] J. Parkka, M. Ermes, P. Korpipaa, J. Mantyjarvi, J. Peltola, and I. Korhonen, “Activity classification using realistic data from wearable sensors,” IEEE Trans. Inf. Technol. Biomed., vol. 10, no. 1, pp. 119–128, Jan. 2006.

Alan Godfrey (M’08) received the B.Eng. degree in electronic engineering and the Ph.D. degree in biomedical electronics from the University of Limerick, Limerick, Ireland, in 2004 and 2008, respectively. After a research assistant position at the University of Limerick, he is now a Postdoctoral Research Fellow with the School of System Engineering at the University of Reading, Berkshire, U.K. His current research interests include human motion analysis and physical activity, signal processing, human–computer interaction, cognitive and nutritional related issues.



Richard Conway (M’01) received the B.Eng. degree in electronic engineering and the Ph.D. degree from the University of Limerick, Limerick, Ireland, in 1990 and 2001, respectively. His Ph.D. research topic was on new residue number system architectures for digital signal processing. His current research includes low power DSP architectures, fault tolerant DSP and signal processing for medical applications. He spent three years with Electrotek Ltd. as a design engineer, before joining the academic staff in the Electronic and Computer Engineering Department, University of Limerick, where he is currently a lecturer, lecturing in digital systems, signal processing architectures, and biometrics.

Maeve Leonard qualified from the Medical Faculty at the National University of Ireland, Galway (NUIG), Ireland, in 1992. She qualified as a Member of the Royal College of Psychiatrists in 1997 and worked at the Queen’s Medical Centre, Nottingham, U.K., as a Specialist Registrar in General Adult and Old Age Psychiatry between 1998 and 2001. She is presently undertaking the M.D. in delirium phenomenology: a longitudinal study of symptom profile in a palliative care population at NUIG. She currently works as a Senior Registrar in Old Age Psychiatry at the Health Service Executive (HSE), Limerick, Ireland, and has a special interest in delirium.

307

David Meagher qualified as a medical doctor in 1988 from University College Dublin (UCD), Dublin, Ireland, and subsequently received the M.Sc. degree in neuroscience from the Institute of Psychiatry, and the M.D. degree from UCD, in 2001, focusing on predictors of outcome in psychotic illness. His current research includes identifying clinical subtypes for delirium and developing electronic methods for early detection of acute cognitive failure. He is a consultant psychiatrist at the Midwestern Regional Hospital, Limerick, Ireland, and Adjunct Professor at the Health Systems Research Centre, University of Limerick.

Gearóid M. Ólaighin received the B.E. degree in electrical engineering and the M.Eng.Sc. degree in microelectronics, both from University College, Cork (UCC), in 1980 and 1982, respectively. He received the Ph.D. degree in biomedical engineering from the National University of Ireland, Galway, in 2000. He spent five years with Integrated Device Technology, Inc., San Jose, CA. He was a Senior Lecturer in Electronic Engineering, Assistant Dean, Research for the College of Informatics and Electronics and founding Director of the Biomedical Electronics Laboratory, University of Limerick. Since joining NUI Galway, he has established the Bioelectronics Research Cluster at the National Centre for Biomedical Engineering Science (NCBES) and is currently leader of that research grouping, Professor of Electronic Engineering and Head of the Department of Electrical and Electronic Engineering at NUI Galway. He is a member of the Editorial Board of Medical Engineering & Physics Prof. Ó Laighin is a Fellow of Engineers Ireland, a Fellow of the Institution of Engineering and Technology (FIET), and Senior Member of the Institute of Electrical and Electronics Engineers (SMIEEE). He is a Fulbright Scholar.
