WO2010139859A1 - Electrocardiography data analysis - Google Patents

Abstract

An apparatus, a method, and a computer program for predicting a risk to a cardiac death are presented. The apparatus (100) comprises a processor (116) configured to calculate morphoiogica! parameter values (values of total cosine R-to-T, values of QRS complex loop asymmetry, values of T wave loop dispersion, values of QRS complex and T wave principal component analysis, values of an angle between a main QRS complex loop vector and a main T wave loop vector) of QRS complex and T wave from a sample period within electrocardiography data recorded from a subject; to calculate heart rate values from the sample period; to form an association result between the change of the morphological parameter values and change of the heart rate values; and to predict a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition.

Description

Electrocardiography data analysis

Field

The invention relates to predicting a risk to a cardiac death on the basis of an electrocardiography data analysis.

βackg round

Heart disease is the leading cause of death in most industrialized nations. Health care is able to treat and manage cardiovascular illnesses in a surprisingly effective fashion. However, the prerequisite is that the persons at risk are identified. Various tests have been developed, but due to the dire consequences of the disease, it is well worth the effort to continue the refinement of the methodology with which the risk to a cardiac death may be assessed.

Brief description The present invention seeks to provide an improved apparatus, an improved method, and an improved computer program for predicting a risk to a cardiac death.

According to an aspect of the present invention, there is provided an apparatus as specified in claim 1. According to another aspect of the present invention, there is provided a method as specified in claim 6.

According to another aspect of the present invention, there is provided a computer program as specified in claim 11.

According to another aspect of the present invention, there is provided a computer program on a carrier as specified in claim 12.

According to another aspect of the present invention, there is provided another apparatus as specified in claim 13. List of drawings

Embodiments of the present invention are described below, by way of example only, with reference to the accompanying drawings, in which

Figure 1 illustrates embodiments of an apparatus; Figures 2, 3, 4 and 5 illustrate processing of electrocardiography data;

Figure 6, 7 and 8 illustrate morphological parameter values in relation to heart rate values;

Figures 9 and 10 illustrate further processing of electrocardiography data;

Figures 11 to 20 illustrate a manuscript of a study appended at the end of the 'Description of embodiments' section; and

Figures 21 to 24 illustrate some further embodiments.

Description of embodiments The following embodiments are exemplary. Although the specification may refer to "an" embodiment in several locations, this does not necessarily mean that each such reference is to the same embodiment(s), or that the feature only applies to a single embodiment. Single features of different embodiments may also be combined to provide other embodiments. Figure 1 illustrates embodiments of an apparatus 100. Figure 1 only shows some elements whose implementation may differ from what is shown. The connections shown in Figure 1 are logical connections; the actual physical connections may be different. Interfaces between the various elements may be implemented with suitable interface technologies, such as a message interface, a method interface, a sub-routine call interface, a block interface, or any means enabling communication between functional sub- units. It should be appreciated that the apparatus 100 may comprise other parts. However, such other parts may be irrelevant to the actual invention and, therefore, they need not be discussed in more detail here. It is also to be noted that although some elements are depicted as separate ones, some of them may be integrated into a single physical element. The specifications of the apparatus 100 may develop rapidly. Such development may require extra changes to an embodiment. Therefore, all words and expressions should be interpreted broadly, and they are intended to illustrate, not to restrict, the embodiments. The apparatus 100 comprises a processor 116. The processor 116 is configured to calculate morphological parameter values of QRS complex and T wave from a sample period within electrocardiography (ECG) data recorded from a subject, and to calculate heart rate values from the sample period. The processor 116 is also configured to form an association result between the change of the morphological parameter values and change of the heart rate values, and to predict a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition. Next, we will describe the structure of the apparatus 100 in more detail, but, upon this, we will describe the processing of the ECG data in more detail. The apparatus 100 may be an electronic digital computer, which may comprise, besides the processor 116, a working memory 112, and a system clock 126. Furthermore, the computer 100 may comprise a number of peripheral devices. In Figure 1 , some peripheral devices are illustrated: a nonvolatile memory 110, an input interface 124, and a user interface 128 (such as a pointing device, a keyboard, a display, etc.). The user interface 128 may be used for user interaction: a physician may view the ECG data with the user interface 128, and a predicted risk to a cardiac death may be shown to the physician with the user interface 128. Naturally, the computer 100 may comprise a number of other peripheral devices, not illustrated here for the sake of clarity.

The system clock 126 constantly generates a stream of electrical pulses, which cause the various transferring operations within the computer to take place in an orderly manner and with specific timing.

Depending on the processing power needed, the computer 100 may comprise several (parallel) processors 116, or the required processing may be distributed amongst a number of computers 100. The computer 100 may be a laptop computer, a personal computer, a server computer, a mainframe computer, or any other suitable computer. As the processing power of portable communications terminals, such as mobile phones, is constantly increasing, the apparatus 100 functionality may be implemented into them as well. Besides the contemporary computers utilizing binary digits, bits, for representing data and instructions by the use of the Binary number system's two-binary digits "0" and "1 ", the emerging quantum computers may also be used, such quantum computers utilizing quantum bits, qubits, instead of bits.

It is to be noted that the apparatus 100 functionality may be implemented, besides in computers, in other suitable data processing equipment as well: in an electrocardiograph, or in an other device recording and/or analyzing ECG data.

The term 'processor' refers to a device that is capable of processing data. The processor 116 may comprise an electronic circuit or electronic circuits implementing the required functionality, and/or a microprocessor or microprocessors running a computer program 104C implementing the required functionality. When designing the implementation, a person skilled in the art will consider the requirements set for the size and power consumption of the apparatus, the necessary processing capacity, production costs, and production volumes, for example. The electronic circuit may comprise logic components, standard integrated circuits, application-specific integrated circuits (ASIC), and/or other suitable electronic structures.

The microprocessor 116 implements functions of a central processing unit (CPU) on an integrated circuit. The CPU 116 is a logic machine executing a computer program 104C, which comprises program instructions 106C. The program instructions 106C may be coded as a computer program using a programming language, which may be a high-level programming language, such as C, or Java, or a low-level programming language, such as a machine language, or an assembler. The CPU 116 may comprise a set of registers 118, an arithmetic logic unit (ALU) 122, and a control unit (CU) 120. The control unit 120 is controlled by a sequence of program instructions 106D transferred to the CPU 116 from the working memory 112. The control unit 120 may contain a number of microinstructions for basic operations. The implementation of the microinstructions may vary, depending on the CPU 116 design. The microprocessor 116 may also have an operating system (a dedicated operating system of an embedded system, or a real-time operating system), which may provide the computer program 104C with system services.

There may be three different types of buses between the working memory 112 and the processor 116: a data bus 130, a control bus 132, and an address bus 134. The control unit 120 uses the control bus 132 to set the working memory 112 in two states, one for writing data into the working memory 112, and the other for reading data from the working memory 112. The control unit 120 uses the address bus 134 to send to the working memory 112 address signals for addressing specified portions of the memory in writing and reading states. The data bus 130 is used to transfer data 114 from the working memory 112 to the processor 116 and from the processor 116 to the working memory 112, and to transfer the instructions 106C from the working memory 112 to the processor 116.

The working memory 112 may be implemented as a random- access memory (RAM), where the information is lost after the power is switched off. The RAM is capable of returning any piece of data in a constant time, regardless of its physical location and whether or not it is related to the previous piece of data. The data may comprise ECG data, any temporary data needed during the analysis, program instructions, morphological parameter values, etc. The non-volatile memory 110 retains the stored information even when not powered. Examples of non-volatile memory include read-only memory (ROM), flash memory, magnetic computer storage devices such as hard disk drives, and optical discs. As is shown in Figure 1 , the non-volatiie memory 110 may store both data 108 and a computer program 104B comprising program instructions 106B.

An embodiment provides a computer program 104A comprising program instructions 106A which, when loaded into the apparatus 100, cause the apparatus 100 to calculate morphoiogica! parameter values of QRS complex and T wave from a sample period within electrocardiography data recorded from a subject, to calculate heart rate values from the sample period, to form an association result between the change of the morphological parameter values and change of the heart rate values, and to predict a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition.

The computer program 104A may be in source code form, object code form, or in some intermediate form. The computer program 104A may be stored in a carrier 102, which may be any entity or device capable of carrying the program to the apparatus 100. The carrier 102 may be implemented as follows, for example: the computer program 104A may be embodied on a record medium, stored in a computer memory, embodied in a read-only memory, carried on an electrical carrier signal, carried on a telecommunications signal, and/or embodied on a software distribution medium. In some jurisdictions, depending on the legislation and the patent practice, the carrier 102 may not be the telecommunications signal.

Figure 1 illustrates that the carrier 102 may be coupled with the apparatus 100, whereupon the program 104A comprising the program instructions 106A is transferred into the non-volatile memory 110 of the apparatus. The program 104B with its program instructions 106B may be loaded from the non-volatile memory 110 into the working memory 112. During running of the program 104C, the program instructions 106C are transferred via the data bus from the working memory 112 into the control unit 120, wherein usually a portion of the instructions 106D resides and controls the operation of the apparatus 100.

There are many ways to structure the program 104A/104B/104C. The operations of the program may be divided into functional modules, subroutines, methods, classes, objects, applets, macros, etc., depending on the software design methodology and the programming language used. In modern programming environments, there are software libraries, i.e. compilations of ready made functions, which may be utilized by the program for performing a wide variety of standard operations.

The computer program 104A₁ 104B, 104C, 104D may comprise four separate functional entities (which may be divided into modules, subroutines, methods, classes, objects, applets, macros, etc.):

- a first entity to calculate morphological parameter values of QRS complex and T wave from a sample period within electrocardiography data recorded from a subject;

- a second entity to calculate heart rate values from the sample period;

- a third entity to form an association result between the change of the morphological parameter values and change of the heart rate values; and

- a fourth entity to predict a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition. Basically, the software filters input data, performs the principal component analysis to produce vectorcardiograph ic representation of the data, and computes various morphological descriptors of the data on a beat- to-beat basis. The beat-to-beat morphological descriptors may then be correlated to the corresponding beat intervals in order to derive the new index for the data.

Besides these basic entities, there may be a number of other, supplementary entities. Data 114, which comprises electrocardiography data recorded from a subject, may be brought into the working memory 112 via the non-volatiie memory 110 or via the input interface 124. For this operation, there may exist a further software entity. The data 108 may have been brought into the non-voiatile memory 110 via a memory device (such as a memory card, an optical disk, or any other suitable non-volatile memory device) or via a telecommunications connection (via Internet, or another wired/wireless connection). The input interface 124 may be a suitable communication bus, such as USB (Universal Serial Bus) or some other serial/parallel bus, operating in a wireless/wired fashion. The input interface 124 may be directly coupled with an electronic system (electrocardiograph) recording electrocardiography, from a subject via skin electrodes, or there may be a telecommunications connection between the input interface 124 and the electronic electrocardiography recording system. A wireless connection may be implemented with a wireless transceiver operating according to the GSM (Global System for Mobile Communications), WCDMA (Wideband Code Division Multiple Access), WLAN (Wireless Local Area Network) or Bluetooth® standard, or any other suitable standard/non-standard wireless communication means.

To conclude, the apparatus 100 is capable of analyzing electrocardiography data in real-time, i.e. during the recording, or in non-realtime, i.e. any time after completing the recording, and the electrocardiography data may be brought into the apparatus 100 by any means for transferring data.

Next, referring to Figures 2, 3, 4 and 5 processing of electrocardiography (ECG) data is explained in more detail. Usually ten skin electrodes are placed on the subject: one electrode for each wrist, one electrode for each ankle, and six electrodes for the chest. Wires connecting the electrodes to the electrocardiograph may be called leads, but we use the more common definition: a 'lead' refers to a combination of electrodes that form an imaginary line in the body along which the electronic signal are measured. In this discussion, we refer to a 12-iead electrocardiograph, but the embodiments are not necessarily restricted to ECG data produced with such an electrocardiograph.

Figure 2 illustrates eight independent leads of the 12-lead ECG. In order to calculate the morphological descriptors, singular value decomposition may be applied to the eight algebraically independent leads I, II, V1-V6 of the 12-lead ECG sample. Leads I and Il may be called limb leads, and leads V1- V6 may be called precordial leads.

The leads may be reconstructed in an orthogonal lead system so that the first lead contains the maximum energy in one direction and the second the maximum energy perpendicular to the first lead, etc. Figure 3 shows the reconstructed, orthogonal leads SrSe- It has been shown that the first three components Si, S₂, S₃ of the decomposition contain over 99% of the ECG's energy; they are referred to as the dipolar components, which are related to the X, Y and Z leads of the Frank's vectorcardiographic lead system. Figure 4 shows the three- dimensional representation 400 of the QRS and T wave loops and their projections 402, 404, 406 on the main planes. if M is an 8 x n matrix, where each row corresponds to an ECG lead and n represents the number of samples, then there exist two orthogonal matrices: L - [I₁, I₂,... , l₈] e Θ^8xS and R = [n, r₂,... , r₈]e θ^nxn so that the SVD of M₁ Σ, equals to Σ = L^TMR = diag(σi, σ₂... , σ₈)e θ^8x8, where σi>σ₂>...>σ₈≥0. Columns L and R are the left and right singular vectors, respectively, and σι are the singular values of M.

The three-dimensional minimum subspace is spanned by the columns of L^8x3, and P is the projection of matrix M onto L, P = L^TM. Let P_QRS and Pj represent the projections of QRS and T-wave loops onto L, respectively. Unit vector UT represents the main orientation of the T-wave loop (P_T) in the three-dimensional space. The TCRT is the average of the cosines of the angles between the UT and each vector, PQRSO, i = ΪQRS start -> ΪQRS end), of the QRS complex whose resultant exceeds 50% of the maximum value of the resultant E3D in Figure 5. Besides being 50 %, the predetermined correlation threshold value may also have another suitable value (calibrated by further studies). Another suitable value may be more than 50%: 70%, for example.

In Figure 5, the ECG's energy (resultant) E₃₀ is calculated based on the dipolar components S₁-S₃.

The correlation coefficient (between TCRT and RR, denoted here with TCRT-RR) may be calculated from the data gathered from the beginning of the exercise to the maximum heart rate. The correlation coefficient representing recovery may be calculated from the samples collected within the first three minutes of recovery, excluding the sampie from the maximum heart rate.

TCRT- RR

As expiained above, the morphological parameter values may be values of total cosine R-to-T (TCRT). Besides this, other values may also be utilized: values of QRS complex loop asymmetry, values of T wave loop dispersion, values of QRS complex and T wave principai component analysis, values of an angle between a main QRS complex loop vector and a main T wave loop vector.

As explained above, the sample period may comprise a (post- exercise) recovery period of the subject. The sample period may also comprise sub-sample periods within the electrocardiography data.

As explained above, the processor 1 16 may further be configured to form the comparison result by forming a correlation value between the change of the morphological parameter values and change of the heart rate values. In an embodiment, the processor 1 16 is further configured to analyze the way the comparison result fulfils the predetermined condition by comparing the correlation value to a predetermined correlation threshold value.

The Figure 6 shows a typical TCRT-HR curvature of a normal healthy subject. Figures 7 and 8 on the other hand illustrate the TCRT-HR profile of cardiac patients. In healthy adults the correlation coefficient (TCRT- RR) is generally positive (above 0.5) during exercise. During recovery the correlation is typically lower than during exercise, in seriously ill cardiac patients (patients who have suffered previous myocardial infarction (Ml) or have severe coronary artery disease (CAD)), the correlation coefficients might be negative during exercise and recovery as in Figures 7 and 8. TCRT-HR profile of Figure 7 was taken from a coronary artery disease patient who had suffered a myocardial infarction. TCRT-HR profile of Figure 8 was taken from a (now deceased) CAD patient who had suffered a myocardial infarction. Next, with reference to Figures 9 and 10, further processing of ECG data is explained. The processing is described in the form of a method. The method clarifies and expands the operation of the apparatus 100 described earlier. The method may be implemented as the apparatus 100 or the computer program 104A comprising program instructions 106A which, when loaded into the apparatus 100, cause the apparatus 100 to perform the process to be described. The embodiments of the apparatus 100 may also be used to enhance the method, and, correspondingly, the embodiments of the method may be used to enhance the apparatus 100.

The steps are in no absolute chronological order, and some of the steps may be performed simultaneousiy or in an order differing from the given one. Other functions can also be executed between the steps or within the steps and other data exchanged between the steps. Some of the steps or part of the steps may also be left out or replaced by a corresponding step or part of the step. In 900, raw ECG data (sampling rate 500 Hz) is obtained.

The raw ECG data is processed in two branches: 902-904-906, and 908-910-912-914-916-918.

In 902, the raw ECG data is notch filtered (50 Hz), whereupon, in 904, it is further low-pass filtered (40 Hz)₁ and, finally, in 906, a cubic spline interpolation is applied (through PQ-intervals).

In 908, the raw ECG data is band-pass filtered (0.5-35 Hz), whereupon, in 910, it is again band-pass filtered (15-40 Hz). In 912, R-peak is detected (by template matching). In 916, ectopic beats and abnormally shaped beats are removed. In 916, heart rate and a median filtered heart rate are calculated (with a window where n=51). In 918, sample point extraction is performed. In 920, the two branches are synchronized, and a sample sequence consisting of ten consecutive beats is formed for each sample point

(i).

In 922, T-waves (of the sample sequence) are filtered with a low- pass filter (12-15 Hz).

In 924, representative beats (for leads I₁ Ii, and V1Λ/6) are formed. Furthermore, the sample rate is interpolated from 500 Hz to 1428 Hz.

In 926, QRS boundaries and T wave offset are defined (possibly in a manual fashion). In 928, TCRT and other morphological parameters are calculated.

In 930, it is tested whether there are enough sample sequences: if not, 920 is re-entered, else, 932 is entered.

In 932, correlation coefficients are calculated for exercise and recovery periods. In 934, the data is stored, and, finally, in 936, a risk to a cardiac death of the subject is predicted on the basis of the analysis.

If we compare the sequence of Figure 9, with the core functionality of the above-described apparatus 100, we may identify the following operations (also claimed in the appended claims): in 928, calculate morphological parameter values of QRS complex and T wave from a sample period within electrocardiography data recorded from a subject; in 916, calculate heart rate values from the sample period; in 932, form an association result between the change of the morphological parameter values and change of the heart rate values; and in 936, predict a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition.

In the sequence of Figure 9, the morphological parameter values comprise at least one of the following: values of total cosine R-to-T, values of QRS complex loop asymmetry, values of T wave loop dispersion, values of QRS complex and T wave principal component analysis. As illustrated in Figure 9 by steps 918, 920 and 930, the sample period may comprise a (post-exercise) recovery period of the subject, and/or sub-sample periods within the electrocardiography data.

In 932, the comparison result may be formed by forming a correlation value between the change of the morphological parameter values and change of the heart rate values. in 936, the way the comparison result fulfils the predetermined condition may be analyzed by comparing the correlation value to a predetermined correlation threshold value. Figure 10 illustrates in more detai! the calculation of main variables.

In 1000, input is obtained: a signal (including leads I₁ H₁ and V1-V6), QRS start and end, T-wave end, Signal Fs = 1428 Hz.

In 1002, decomposition of signal is performed (SVD = Singular Value Decomposition), In 1004, the signal is normalized. In 1006, DC- component is removed from the signal.

In 1008, the resultant (E₃₀) is calculated. In 1010, R-peak and T- peak are detected. In 1012, QRS and T wave loops are formed.

In 1014, QRS vectors whose resultant exceeds the threshold are extracted. In 1016, T loop orientation is determined, in 1018, TCRT is calculated, In 1020, QRS-T angle is calculated.

In 1022, SVD is performed separately for QRS and T wave to acquire the singular values for the calculation of the PCA ratios. In 1024, PCA1-3 ratios for QRS and T are calculated.

Asymmetry of QRS complex loop may be used as a morphological parameter to replace TCRT as an interpreter of the recovery phase of the exercise test. The symmetry measure of the loop structure is calculated as follows. First, the three-channel signal is resampled equidistantly with respect to the arc-length. Second, the signal is centered by removing the mean. Third, a local coordinate system is established according to the behaviour of the signal in time. This is accomplished by estimating three natural directional vectors: - the "forward-pointing" vector towards the apex of the loop structure (e.g. towards R-peak),

- the "left-pointing" average vector from the beginning to the apex of the loop, and - the "up-pointing" cross-product vector orthogonal to the previous ones.

Fourth, the vectors are then normalized and collected as column vectors into a 3-by-3 matrix A. Finally, the symmetry measure is calculated as the base-10 logarithm of the condition number of the matrix A using the spectral matrix norm induced by the L2 vector norm.

The symmetry measure is non-negative and describes the collinearity of the directional vectors which in turn describe the form of the loop structure. When the directional vectors form an orthonormal basis, the symmetry measure is zero. Furthermore, when the forward and the left pointing vectors are not orthogonal, the symmetry measure is positive.

For example, a perfectly elliptical loop structure would results in a symmetry measure of zero. Now, if a local imperfection was added to the loop structure by buckling it, the left-pointing directional vector would shift accordingly because the length of the loop structure would change. In other words, the equidistant resampling would assign more points to the buckled area shifting the left-pointing vector towards it, which, in turn, would make the lines in the directions of the forward and left pointing vectors more collinear. Naturally, this will lead to an increase in the value of the symmetry measure indicating increased asymmetry. Besides the above-described values of QRS complex loop asymmetry, also other suitable values relating to the symmetry/asymmetry measure of QRS loop, R wave, and/or QRS complex may be used as the morphological parameter values. in an embodiment, two or more associations may be used for the final prediction. In an embodiment, morphological parameter values of QRS complex and T wave are calculated separately from two sample periods, i.e. from an exercise period and from a (post-exercise) recovery period, within electrocardiography data recorded from a subject, heart rate vaiues are calculated from the exercise period and from the recovery period, association results are calculated between the change of the morphologica! parameter values and change of the heart rate values for the exercise period and for the recovery period, a combined association result is formed from the association result calculated for the exercise period and from the association result calculated for the recovery period, and the risk to the cardiac death of the subject is predicted on the basis of the way the combined association result fulfils a predetermined condition for the combined association result. Figure 21 illustrates the TCRT rate-relation observed in three different test-subjects: a subject who was alive after the follow-up, and two patients who both died suddenly during the follow up of this study. The beat- to-beat TCRT-RR patterns of three different patients during exercise (left coiumn) and recovery (right column) are illustrated. Panel A illustrates a typical TCRT-RR pattern in a patient who remained alive during the follow-up. Panels B and C illustrate the TCRT-RR patterns of two CAD patients who both suffered sudden cardiac death during the follow-up of the study. The TCRT- RR relation in panel B is inverted from that of a norma! healthy subject during exercise, and therefore the TCRT-RR index is negative. In panel C, the TCRT- RR relation is negative during both exercise and recovery.

Figure 23 shows the unadjusted and adjusted hazard ratios and their corresponding 95 % confidence intervals (Cl's) for the optimally dichotomized variables in the prediction of cardiac and sudden cardiac death. Continuous parameters were dichotomized according to their optimum cut-off points shown within the parenthesis. Dichotomized parameters were entered one at a time into a stepwise backward model with following clinical risk markers: age, beta blockers (yes/no), body mass index, diagnoses of coronary artery disease (yes/no), diabetes (yes/no), left ventricular ejection fraction, maximum heart rate, metabolic equivalent, previous infarction (yes/no), sex (male/female) and smoking (yes/no). Cl = confidence interval; COMBINED 1 = exercise TCRT-RR or recovery TCRT-RR below optimum cut-off point; COMBINED 2 - both exercise TCRT-RR and recovery TCRT-RR below optimum cut-off point; EX = exercise period; QRST-RR = correlation coefficient between measured QRS/T angles and corresponding RR intervals; REC = recovery period; TCRT-RR = correlation coefficient between measured TCRT values and corresponding RR intervals. Combined 0: TCRT-RR_e* >= threshold AND TCRT-RR_rec >= threshold; Combined 1 : TCRT-RR_6x < threshold OR TCRT-RR_rec < threshold; and Combined 2: TCRT-RR_6x < threshold AND TCRT-RR_rec < threshold; wherein threshold = optimum cut-off point (in ROC-curve). Reduced recovery TCRT-RR (unadjusted hazard ratio of 5.50; 95% CI: 2.83 - 10.69, p<0.001 ) and exercise TCRT-RR (unadjusted hazard ratio of 4.25; 95% Cl: 2.17 - 8.29, ρ<0.001) values were strong predictors of cardiac death. A new categorical variable was constructed as a combination of these two variables (named COMBINED in Figures 23 and 24): patients were divided into three groups based on their exercise and recovery TCRT-RR values. Group 1 consisted of patients with good rate-dependency indexes (good meaning TCRT-RR_Eχ and TCRT-RRRE_C values above the optimum cut-off points, see Figure 23); Group 2 consisted of patients with poor rate-dependence during exercise or recovery; and Group 3 consisted of patients with poor rate-dependence during both exercise and recovery. The unadjusted risk of cardiac death in Group 3 was 9.5 times higher (95% Cl: 4.1 - 22.2; pθ.001 , Figure 22) than in Group 1. For sudden cardiac death, the risk was 11.4 times higher in Group 3 (95% Cl: 4.1 - 32.0; p<0.001 , Figure 22) as opposed to Group 1. In Figure 22, Kaplan-Meier curves for cardiac and sudden cardiac death in patients divided into three groups: Group 1 consisted of patients with exercise TCRT-RR and recovery TCRT-RR above the optimum cut-off points; Group 2 consisted of patients with exercise TCRT-RR or recovery TCRT-RR below the optimum cut-off point; Group 3 consisted of patients with poor exercise and recovery rate-dependence of TCRT.

The sensitivity and specificity, as well as the positive and negative predictive values for the new risk markers are presented in Figure 24. In Figure 24: *p < 0.05 and f p<0.01 ; AUC = area under the receiver operating characteristics curve; COMBINED 1 = exercise TCRT-RR or recovery TCRT- RR below optimum cut-off point; COMBINED 2 = both exercise TCRT-RR and recovery TCRT-RR below optimum cut-off point; EX = exercise; NPV = negative predictive value; PPV = positive predictive value; QRST-RR = correlation coefficient between QRS/T angle and RR interval; REC = recovery; TCRT-RR = correlation coefficient between TCRT values and RR intervals; HRR - one minute heart rate recovery.

We will conclude with a thorough manuscript of a study including examples and statistical analysis of the results. This discussion is not intended to restrict the embodiments, but to serve as a tool offering a deeper understanding of the subject matter, and even offering further definitions for the embodiments. The following description of background art may include insights, discoveries, understandings or disclosures, or associations together with disclosures not known to the relevant art prior to the present invention but provided by the invention. Some such contributions of the invention may be specifically pointed out below, whereas other such contributions of the invention will be apparent from their context. ABSTRACT

Analysis of the spatial characteristics of QRS-complex and T-wave morphology from standard 12-Iead electrocardiogram (ECG) has gained recent clinical interest. However, knowledge about the dynamics and rate- dependence of the morphological patterns of the QRS complex and T-wave is sparse. We assessed the rate-dependence and dynamics of the morphological patterns of the T-wave and QRS-T-wave relationship from 40 exercise ECG recordings: 21 coronary artery disease (CAD) patients and 19 non-CAD subjects. Several morphological descriptors, such as the total cosine R to T (TCRT), T-wave loop dispersion (TWLD) and principal component ratios (PCA-ratio) were analyzed at different sample points throughout the measurement. Their rate-dependence was assessed in individual subjects and in the pooled data (CAD + non-CAD). In individual subjects, TCRT displayed significant linear relation to RR-intervals: median R² during exercise was 0.82 [0.58, 0.94; 25^th and 75^th percentile, respectively] and during recovery 0.77 [0.56, 0,89], Individual rate-dependence in the other parameters was weak and often characterized with large inter-individual variation. Evident hysteresis in the values of the morphological descriptors was observed at the onset and more prominently immediately after the cessation of exercise. Significant linear rate-dependence exists in the morphological patterns of cardiac repolarization and in the relationship between depolarization and repolarization characterized by TCRT in individual subjects, LIST OF ABBREVIATIONS bpm beats per minute

CAD Coronary artery disease

ECG Electrocardio/ -gram, -graphy

GEE Generalized estimation equation

HR Heart rate

Hz Hertz

LBBB Left bundle branch block

Ml Myocardial infarction

PCA Principal component analysis

PCAi 0 = 1 , 2, 3) Principal component ratio

RBBB Right bundle branch block

RR R-R - interval

SCD Sudden cardiac death

SD Standard deviation

SVD Singular value decomposition

SEM Standard error of mean

TCRT Total cosine R to T

TWLD T-wave loop dispersion

TWR T-wave residuum

W Watt

VT Ventricular tachycardia

1. INTRODUCTION

Abnormal cardiac depolarization and repolarization, measured from the standard 12-lead electrocardiogram (ECG), have been associated with adverse clinical outcome in several populations [1 , 2]. The traditional methods in the analysis of these abnormalities have been the measurement of duration of the QRS complex and the QT-interval. Methodological problems related to the measurement of the QT-interval as a marker of repolarization are its strong rate-dependence, inaccuracy of rate-correction formulas and problems related to the exact definition of the end of the T-wave. In addition, measurement of the duration of the depolarization and repolarization may yield a limited view of the electrogenesis of ventricular activation and deactivation.

New descriptors of morphological patterns have been proposed in attempts to improve the characterization of depolarization and repolarization phenomena as well as improving the risk stratification of patients [3-7]. Some of these new descriptors have been shown to possess clinical value, such as the total cosine R to T (TCRT) [5-7], which describes the spatial angle between the main vectors of ventricular depolarization and repolarization wavefronts, the T-wave loop dispersion (TWLD) [6], which reflects the variability of the T-wave vector loop and the T-wave complexity described with principal component ratio (PCA-ratio) [4].

Descriptors of QRS and T-wave morphology have usually been measured from a single QRS-T complex of the standard 12-lead ECG [3-5, 7- 11], or from averaged complexes [12, 13]. However, little is known about the dynamics and rate-dependence of these morphological variables. One previous study [13], assessing the circadian dynamics of TCRT, T-wave residuum (TWR) and corrected QT-intervals from 24-hour Holter recordings showed that TCRT decreased at higher heart rates. However, the dynamic range of heart rate is limited in standard 24-hour Holter recordings; therefore, we wanted to explore the dynamics and rate-dependence in individual subjects and at group level during a standard exercise stress test, where the dynamic range of heart rate is greater. Additionally, we aimed to examine the behaviour of the new morphological descriptors in patients with coronary artery disease (CAD) and compare these with the corresponding values from patients without CAD (non-CAD). As far as we are aware, this is the first study to assess the dynamics of these morphological descriptors from the exercise ECG. 2. METHODS

Population

A total of 55 exercise ECG recordings were selected for this study. The study population consisted of two subgroups: Thirty-two subjects with proven coronary artery disease (CAD), defined as prior Ml and ≥50% stenosis in one or more coronary arteries in coronary angiography, and 23 patients without any history of Ml or significant stenosis in the coronary angiography. Subjects with left or right bundle branch block (LBBB, RBBB, respectively) were excluded from the study since RBBB or LBBB have a major impact on the TCRT values. In addition, CAD patients with inverted T-waves during baseline ECG were excluded as the inverted T-waves in general switched polarity during exercise resulting in a negative correlation between measured TCRT and RR -intervals. After exclusion of ineligible subjects, the pilot population consisted of 21 CAD patients and 19 patients without CAD. The characteristics of these two groups are listed in Figure 17. Exercise Test Protocol

Symptom limited exercise tests were performed using a bicycle ergometer. The Mason-Likar modification with two additional leads was used in the 12-lead ECG recording. The exercise protocol consisted of an initial workload of 20 - 3OW with a gradual increase of load in steps of 10 - 3OW per minute. Continuous digital ECG at 500 Hz was recorded throughout the test with CardioSoft exercise ECG system (Version 4, 14, GE Medical Systems, Freiburg, Germany).

Signal Processing and Sample Extraction The signal processing and the subsequent analysis were carried out with software written in Matlab (Matlab v. 7.5, MathWorks, inc. Natick, MA). Each ECG recording was processed with signal processing methods to suppress the noise present in the recordings, Powerline interference was reduced with a notch filter at 50 Hz. Baseline wander was removed with cubic spline interpolation. High frequency noise was suppressed with a 40 Hz lowpass filter. Ectopic and abnormally shaped beats were removed from the analysis. Samples consisting of 10 consecutive beats were extracted from different points of the measurement. The first sample was taken before the beginning of exercise; this sample was considered to represent the baseline ECG. The following samples were taken at different heart rates (e.g. 90bpm, l OObpm, 110bpm, and so forth) up to the maximum heart rate. Samples during the recovery period were taken in the same manner. In addition, three samples were taken during the recovery period at 30, 60 and 180 seconds after the maximum heart rate. Each sample was visually inspected; noisy or otherwise invalid samples were discarded and replaced with more suitable samples. Representative median beats of each independent ECG lead (I₁ Ii₁ V1-V6) were constructed from each accepted sample to further reduce the noise.

Analysis of T-wave Morphology Descriptors

Analysis of the representative beats was performed with custom- made software validated in previous studies [6, 14, 15]. The software performs singular value decomposition (SVD) on the ECG. Several QRS and T-wave loop parameters are then automatically calculated based on the decomposition. The TCRT (Figure 11) is computed from the first three components and it is the cosine of the angles between the main vectors of the QRS and T-wave loop (see the appendix for further detail). The TCRT receives values between -1 and 1 , where the value -1 equals to 180° deviation between the main vectors of ventricular depolarization and repolarization wavefronts and a value of 1 is equal to a deviation of 0°.

The spatial irregularities of QRS and T-wave loop are measured with loop dispersion by adjusting a rectangle around the loop in a two- dimensional plane spanned by the two largest components of the SVD decomposition and dividing it to 100 (10x10) sub-rectangles (Figure 11 ). The number of sub-rectangles that are passed through by the tip of the cardiac vector during one cardiac cycle defines the dispersion of QRS and T-wave loop, QRSLD and TWLD, respectively. In addition, the width and the height of the T-wave loop are calculated. The long axis of the loop defines its width and the longest perpendicular axis defines its height. The principal shape and the complexity of the T-wave vector loop were assessed with three different PCA ratios: PCA1 -3 (see the appendix for further detail).

Linear regression was used to examine the rate-dependence and dynamics of the morphological descriptors in individual subjects during exercise and recovery. The correlation coefficient, R², slope and the intercept were calculated from the data that was gathered from the beginning of the exercise to the maximum heart rate. The coefficients representing recovery were calculated from the samples collected within the first three minutes of recovery, excluding the sample from the maximum heart rate. Statistical analysis

The data was analyzed using the SPSS version 15 (SPSS, Inc., Chicago, Illinois). The normality of quantitative variables was tested with Shapiro-Wiikes test. To account for correlations between repeated measurements, linear regression with generalized estimation equations (GEE) [16] was used to examine the change in the observed morphological values during exercise and recovery and to assess the differences between the two sub-groups. Post-hoc comparisons were performed with Mann-Whitney U test for unequally distributed data and Student's t-test for data with a normal distribution and they were corrected with Bonferroni correction in order to compensate for the multiple tests. All tests were two-sided and p value < 0.05 was considered statistically significant. The continuous values are presented as mean ± standard deviation unless otherwise stated. 3. RESULTS TCRT during Exercise and Recovery At the beginning of the exercise, the TCRT profile was characterized by a short plateau phase, where the observed values followed the previous baseline values for a brief moment, until they started to decrease towards the exercise peak (Figure 12). A similar hysteresis phenomenon (however, longer in duration) was observed after the cessation of exercise. In some subjects, the TCRT values continued to decline for a short period after the exercise peak, until they began to increase. Generally, the minimum TCRT values were achieved at the peak HR or within 30 seconds after it had occurred. The observed TCRT values were generally lower during recovery than during exercise.

In individual subjects, the linear relation between the TCRT and RR-intervals was very strong. The median R square during exercise was 0.82 [0.58, 0.94, 25^th and 75^th percentile, respectively] and during recovery 0.77 [0.56 - 0.89]. The results of the individual linear regression analyses are shown in Figure 18. There were no significant differences in the coefficients between exercise and recovery.

As the dynamics of TCRT was analyzed in the pooled data, TCRT showed marked rate-dependence during exercise and recovery with higher values at lower heart rates (Figure 13A). Comparisons between the subgroups indicated that there was significant difference in the TCRT values during exercise (GEE: p = 0.026) and recovery (GEE: p = 0.032). Subjects in the CAD group received generally lower TCRT values than the non-CAD group throughout the measurement and the difference was significant at 110 bpm and 120 bpm (p < 0.01 and p < 0.02, respectively) during exercise and at 140 bpm, 130 bpm and 120 bpm (p < 0.05 at each rate category) during recovery. The TCRT-HR profile was similar in both groups throughout the measurement. However, the recovery of TCRT, measured at 30, 60 and 180 seconds after the maximum HR, was significantly slower in the CAD group than in the non- CAD group (Figure 14): This difference was statistically significant at 60 and 180 seconds after the maximum heart rate (p = 0.041 and p = 0.001 , respectively).

TWLD during Exercise and Recovery

Figure 13B iliustrates the dynamics of the mean TWLD values during exercise and recovery. The observed values of TWLD decreased during exercise and achieved their minimum value generally at the peak heart rate or slightly thereafter. Despite the increasing trend in the mean TWLD values during recovery, the change was not statistically significant.

In individual subjects, the linear relation between TWLD and RR- intervals was moderate and characterized by large variation between subjects. The TWLD-RR slopes differed significantly between exercise and recovery (p

< 0.01), with greater slopes during exercise (Figure 18).

Comparisons between the sub-groups indicated that the TWLD-HR profile differed significantly during recovery (GEE, p = 0,023). The mean TWLD values in the non-CAD group increased towards the baseline value, whereas the values in CAD remained the same. The difference was significant at 100 and 90 bpm <p - 0.0006 and p = 0.0002, respectively). QRSLD during Exercise and Recovery

Figure 13C shows the dynamics of the mean QRSLD during the measurement. It did not experience significant change during exercise or recovery. The linear relation between QRSLD and RR-intervals in individual subjects was weak and varied greatly between the subjects (Figure 18). In addition, no significant differences existed between the subgroups in any of the QRSLD parameters. T-Wave Width and Height during Exercise and Recovery

Figures 15A and 15B illustrate the mean dynamics of the T-wave's width and height in the pooled data during exercise and recovery. The width of the loop did not experience significant change during exercise; however, there was a significant change after the exercise peak as the width increased notably during early recovery. Figure 15A clearly shows the difference in the values between the exercise and recovery. The maximum width was generally observed within one minute after the maximum HR, after which the width started to progressively decrease towards the baseline value. In individual subjects, the linear relation between T-wave width and RR-intervals was moderate and characterized with large variation (Figure 18).

The change in the height of the T-wave loop was not significant during exercise or during recovery. Nevertheless, there was a weak trend in the values of T loop height during exercise and recovery with higher values at higher heart rates (Figure 15B). In individual subjects, the linear relation between T-wave loop's height and RR-intervals differed between exercise and recovery with greater R square values during exercise (Figure 18). PCA during Exercise and Recovery Figures 16A, 16B₁ and 16C illustrate the dynamics of the PCA ratios of T-wave loop during exercise and recovery. During exercise, PCA1 decreased whereas PCA2 and PCA3 on the other hand increased. The change in the values during exercise was significant in a!l of the ratios (GEE: p = 0.01 , p = 0.027, p = 0.05 for PCA1 , -2 and -3 ratios, respectively). During the recovery period only PCA1 experienced significant change (GEE, p < 0.041). However, all the ratios displayed a notable change in the observed values between exercise and recovery.

Weak rate-dependence existed in the individual PCA values and it was characterized by large variation. There was no significant difference between the subgroups in any of the PCA ratios.

Hysteresis

Figures 12 through 16C reveal that there is evident hysteresis in the values of morphological descriptors, in the sense that the descriptors at a given heart rate during recovery receive different values than during exercise. Hysteresis can be seen in each of the studied parameters; however, the differences in the values between the exercise and recovery are largest in the PCA ratios immediately after the exercise peak.

Correlations between Electrocardiographic measurements The results of the correlations between the electrocardiographic measurements are shown in Figures 19 and 20 for exercise and recovery, respectively. The RR-intervals had a significant positive correlation with TCRT₁ TWLD, and PCA1 and negative correlation with PCA2 and PCA3 during exercise. During recovery, the TWLD didn't correlate with RR-intervals and in the PCA-ratios the correlation was inverted in contrast to exercise. The PCA-ratios correlated significantly between each other. Especially, the correlation between PCA1 and PCA2 ratios was very strong.

4. DISCUSSION

The results emphasize the significant influence of heart rate on the morphological patterns of repolarization and the spatial relationship between depolarization and repolarization (TCRT, TWLD, PCA1-3) in the human heart.

In particular, the TCRT displayed a remarkable rate-dependence in individual subjects during exercise and recovery. The present observation supports the concept that a similar rate-dependency exists in TCRT as that observed in the duration of cardiac repolarization.

Relations to Previous Studies The rate-dependence of TCRT has been previously reported by

Smetana et al [13]. Their study indicated that TCRT achieves lower values at higher heart rates. The results from our study concur with this observation. However, the magnitude of individual TCRT-RR relation shown in our study was much greater than that reported by Smetana et al [12] [(women: r = 0.464+0.21 men: r = 0.549+0.21) vs. present study during exercise 0.928 (0.542 - 0.984) and recovery 0.932 (0.610 - 0.979)], which might be explained by the physical stress and broader dynamic range of the exercise ECG. Despite the strong individual TCRT-RR correlations, the rate-dependence in the pooled data was only moderate (Figures 19 and 20). This is attributable to the large inter-individual variation in the measured values of TCRT. Due to this inter-individual variation, the rate-correction of TCRT might be difficult to perform satisfactorily with a generalized correction formula, but individualized rate-correction should be considered instead. Similar inter-individual variations have been recently described in the rate-dependence of QT interval duration. As far as we are aware, there are no previous reports on the rate- dependence of TWLD or PCA ratios. Nevertheless, previous studies have quantified the changes in T-wave shape during and after exercise in normal subjects. Simoons and Hugenholtz [17] reported gradually decreasing T-wave amplitudes during exercise, which can be seen in the observed values of PCA1 and T-wave loop width reported in this study. Our study revealed that the PCA1 values decrease progressively towards the exercise peak, which concurs with the observations made by Simoons and Hugenholtz. Langley et al [18] who studied the changes in T-wave shape following exercise with T- wave amplitude and two indices of T-wave symmetry. They reported that the repolarization after exercise was characterized by tall and symmetric T-waves. This observation is in line with the results of our study: PCA1 increased rapidly after the exercise, this being evident in the increase of T-wave amplitude. In addition, the width of thθ T-wave loop, which partly reflects the amplitude of the T-wave, increased right after the peak exercise indicating increase in the amplitude of T-wave. Furthermore, both PCA2 and PCA3 ratios decreased immediately after the exercise peak, reflecting a rapid change in the distribution of ECG's energy among the reconstructed leads, and thus in the morphology of the T-wave loop.

Hysteresis was observed in the analyzed morphological descriptors. It was more obvious in the PCA-ratios, which can be explained by the rapid change in the T-wave loop morphology immediately after the exercise peak, as explained above. Surprisingly, the hysteresis seemed also to exist in TCRT and TWLD. Similar phenomenon in cardiac repolarization has been observed earlier in RR-QT relationship [19, 20] as well as in the dynamics of action potential duration during cardiac pacing [21], Relationship between Morphological Parameters The TCRT correlated significantly with TWLD, T wave loop height and with the PCA-ratios during exercise (Figure 19). During recovery, the correlation between TWLD and PCA1 and PCA2 ratios vanished (Figure 20). This might be due to the change in the T wave loops morphology during recovery. The T-wave parameters reflect the change in the T wave loop better than the TCRT as it takes only the maximum vector of the T wave loop into account (see the Appendix). TWLD correlated with the PCA ratios (PCA1 and PCA2) and the T wave height as expected; however, the correlation with T wave loop's width was surprisingly not significant. The PCA ratios experienced a significant change between during early recovery (Figures 16A, 16B, and 16C), which contributes to the inverted correlation between RR-intervals and the PCA ratios.

Limitations

This study has taken a step in the direction of defining the rate dependence and dynamics of new morphological descriptors during exercise ECG. However, a few methodological points deserve some comment. The reliability of T-wave morphology analysis may sometimes be questionable due to the extensive noise present in an exercise test. Noise affects the T-wave loop parameters by increasing their values. The effects of noise can be reduced by effective filtering and with the use of averaged representative beats, in addition, our previous studies have shown that TCRT is inherently robust, as the phenomenon it describes is large and therefore less affected by noise. [22, 23]

The advantage of the new morphological descriptors over the QT dispersion is the reproducibility of the measurement. In contrast to the measurement of QT dispersion, the analysis of these new morphological descriptors does not require an exact determination of T-wave offset. Furthermore, sources of human errors are minimized with digital ECG and automated computer analysis. [7]

The size of the study population was small and therefore, the results were more prone to be affected by individual values. The statistical power of some analyses was also limited due to the sample size. However, there is a distinct trend in the observed values of the morphological parameters, clearly indicative of the effects of heart rate on the dynamics of these descriptors.

Some of the analyzed parameters seem not to return to their normal, baseline value during recovery. This is due to the sample extraction and limited data: recovery was monitored for five minutes and apart from the samples taken from different heart rates, fixed samples were taken only as far as from three minute recovery. Longer monitoring of recovery would have revealed the return to normal metrics. Implications The present results clearly show that the TCRT value, which has been already used in risk stratification analysis of various populations, exhibits a strong iinear correlation with heart rate, resembling that of the QT-interval. Therefore, in the analysis of this index, a heart rate correction should be considered in attempts to describe more specifically the global repolarization heterogeneity, not only the heart rate. Furthermore, it is possible that the observed association between the reduced TCRT and mortality may partly result from this rate-dependency. A high heart rate resulting in lower TCRT values is a well known risk factor of mortality [24, 25]. However in the present study, TCRT at high heart rates during exercise, and to a lesser extent at baseline, differed between the subjects with and without CAD₁ indicating that this index is altered in disease states, especially at high heart rates. Therefore, future studies should assess the prognostic value of this index, measured both at rest, adjusted for heart rate, and at high heart rates during exercise.

APPENDIX

PCA ratio. PCA ratios provide information on the principal shape and complexity of the loop. The calculation of the ratios (equations 3 - 5) is based on the singular values, σ,, which measure the distribution of the ECG's energy among the columns of L.

PGi = ^¹X lOO (4)

^• ι

PCA₃ = ^x IQO (5)

Singular value σ, is associated with the /:th column of L The computational details of TCRT, TWLD and the PCA-ratios have been

previously described in detail by Acar et al [7].

REFERENCES

1. Pelliccia A, Di Paolo FM, Quattrini FM, Basso C, Culasso F, Popoli G, De Luca R, Spataro A, Biffi A, Thiene G, Maron BJ. Outcomes in athletes with marked ECG repolarization abnormalities. N Engl J Med. 2008; 358: 152-161.

2. Robbins J, Nelson JC₁ Rautaharju PM, Gottdiener JS. The association between the length of the QT interval and mortality in the Cardiovascular Health Study. Am J Med. 2003; 115: 689-694. 3. Priori SG, Mortara DW, Napolitano C, Diehl L₁ Paganiπi V, Cantu

F, Cantu G, Schwartz PJ. Evaluation of the spatial aspects of T-wave complexity in the long-QT syndrome. Circulation. 1997; 96: 3006-3012.

4. Okin PM, Devereux RB₁ Fabsitz RR, Lee ET₁ Galloway JM₁ Howard BV. Principal component analysis of the T wave and prediction of cardiovascular mortality in American Indians: the Strong Heart Study. Circulation. 2002; 105: 714-719.

5. Zabel M, Acar B, Klingenheben T₁ Franz MR₁ Hohnloser SH, Malik M. Analysis of 12-lead T-wave morphology for risk stratification after myocardial infarction. Circulation. 2000; 102: 1252-1257.

6. Perkiomaki JS₁ Hyytinen-Oinas M, Karsikas M, Seppanen T, Hnatkova K, Malik M, Huikuri HV. Usefulness of T-wave loop and QRS complex loop to predict mortality after acute myocardial infarction. Am J Cardiol. 2006; 97: 353-360. 7. Acar B, Yi G₁ Hnatkova K₁ Malik M. Spatial, temporal and wavefront direction characteristics of 12-lead T-wave morphology. Med Biol Eng Comput. 1999; 37: 574-584.

8. Malik M, Hnatkova K, Batchvarov VN. Post infarction risk stratification using the 3-D angie between QRS complex and T-wave vectors. J Electrocardiol. 2004; 37 Suppl: 201-208.

9. Zabel M, Malik M. Predictive value of T-wave morphology variables and QT dispersion for postmyocardial infarction risk assessment. J Electrocardiol. 2001 ; 34 Suppl: 27-35.

10. Zabel M, Malik M₁ Hnatkova K, Papademetriou V, Pittaras A, Fletcher RD₁ Franz MR. Analysis of T-wave morphology from the 12-lead electrocardiogram for prediction of iong-term prognosis in male US veterans. Circulation. 2002; 105: 1066-1070.

11. Hnatkova K, Ryan SJ, Bathen J₁ Acar B, Batchvarov V₁ Hoium HH₁ Malik M. T-wave morphology differences between patients with and without arrhythmic complication of ischemic heart disease. J Electrocardiol. 2001 ; 34 Suppl: 113-117. 12. Smetana P, Batchvarov VN, Hnatkova K, Camm AJ, Malik M.

Sex differences in repolarization homogeneity and its circadian pattern. Am J Physiol Heart Circ Physiol. 2002; 282: H 1889-97.

13. Smetana P₁ Batchvarov VN, Hnatkova K, Camm AJ, Malik M. Ventricular gradient and nondipolar repolarization components increase at higher heart rate. Am J Physiol Heart Circ Physiol. 2004; 286: H131-6.

14. Karsikas M, Perkiomaki JS, Huikuri H, Seppanen T. Influence of size and location of myocardial infarction on TCRT parameter: a computer simulation study. ConfProc IEEE Eng Med Biol Soc. 2007; 2007: 1414-1417. 15. Linna EH, Perkiomaki JS, Karsikas M, Seppanen T, Savolainen

M, Kesaniemi YA, Makikalϋo T, Huikuri HV. Functional significance of KCNH2 (HERG) K897T polymorphism for cardiac repolarization assessed by analysis of T-wave morphology. Ann Noninvasive Electrocardiol. 2006; 11 : 57-62.

16. Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous outcomes. Biometrics. 1986; 42: 121-130.

17. Simoons ML, Hugenholtz PG. Gradual changes of ECG waveform during and after exercise in normal subjects. Circulation. 1975; 52: 570-577.

18. Langiey P, Di Bernardo D, Murray A. Quantification of T wave shape changes following exercise. Pacing Clin Electrophysiol. 2002; 25: 1230-

1234.

19. Sarma JS, Venkataraman SK, Samant DR, Gadgil U. Hysteresis in the human RR-QT relationship during exercise and recovery. Pacing Clin Electrophysioi. 1987; 10: 485-491. 20. Arnold L, Page J, Attweli D, Cannell M, Eisner DA. The dependence on heart rate of the human ventricular action potential duration. Cardiovasc Res. 1982; 16: 547-551.

21. Patwardhan A, Moghe S. Novel feedback based stimulation protocol shows hysteresis in cardiac action potential duration restitution. Biomed Sci Instrum. 2001 ; 37: 505-510. 22. Lehtola L, Karsikas M₁ Koskinen M, Huikuri H, Seppanen T.

Effects of noise and filtering on SVD-based morphological parameters of the T wave in the ECG. J Med Eng Technol. 2008; 32: 400-407.

23. Karsikas M, Huikuri H, Perkiomaki JS, Lehtola L, Seppanen T. influence of paper electrocardiogram digitizing on T wave and QRS complex morphology parameters. Ann Noninvasive Etectrocardiol. 2007; 12; 282-290.

24. Diaz A, Bourassa IvIG₁ Guertin MC₁ Tardif JC. Long-term prognostic value of resting heart rate in patients with suspected or proven coronary artery disease. Eur Heart J. 2005; 26: 967-974. 25. Reunanen A₁ Karjalainen J₁ Ristola P, Heliovaara M, Knekt P₁

Aromaa A. Heart rate and mortality. J Intern Med. 2000; 247: 231-239.

26. Acar B, Koymen H. SVD-based on-line exercise ECG signal orthogonalization. IEEE Trans Biomed Eng. 1999; 46: 311-321. FIGURE LEGENDS Figure 11 : Three-dimensional QRS and T-wave loops (on the left) and the projection of the T-wave loop onto a two-dimensional plane (on the right). The angle between the main vectors (depicted with arrows) of the two loops determines the TCRT. The XY-projection of the T-wave loop is encompassed with a rectangle, which is divided into 100 (10x10) sub- rectangles. The dispersion of the loop is determined by the number of cells that are passed through by the tip of the T-wave vector during its time-course. Figure modified from [5],

Figure 12: A typical TCRT-HR profile taken from a healthy subject. There is a short plateau at the beginning of the exercise. TCRT receives generally lower values during recovery than during exercise. The first two recovery values (30s recovery [1] and 60s recovery [2]) in this individual are slightly lower than the TCRT values at the corresponding heart rates during exercise.

Figures 13A₁ 13B, and 13C: Dynamics of TCRT (A), TWLD (B) and QRSLD (C) during exercise and recovery. Values from (fixed sample points) baseline (BL), peak exercise (max) and 30, 60 and 180 second recovery values are indicated with (α). Significant difference (p < 0.05) between fixed sample point value and baseline is marked with *. Values are presented as mean ± SEM.

Figure 14: TCRT values at maximum heart rate and 30, 60 and 180 s after the peak heart rate. Significant difference between the groups is marked with ^* (p < 0.05). Values are presented as mean ± SEM.

Figures 15A and 15B: Dynamics of T-wave loops width (A) and height (B) during exercise and recovery. Note the difference in T-wave loops width between the exercise and recovery period. Values from (fixed sample points) baseline (BL)₁ peak exercise (max) and 30, 60 and 180 second recovery values are indicated with (a). Significant difference (p < 0.05) between a fixed sample point and baseline value is marked with ^*. Values are presented as mean ± SEM.

Figures 16A₁ 16B₁ and 16C: Dynamics of TPCA1 (A)₁ TPCA2 (B) and TPCA3 (C) during exercise and recovery. Values from (fixed sample points) baseline (BL)₁ peak exercise (max) and 30, 60 and 180 second recovery values are indicated with (G). Significant difference (p < 0.05) between a fixed sample point and baseline value is marked with a *. Values are presented as mean ± SEM.

Figure 17: Characteristics of the study population. Values are mean ± SD. bpm = beats per minute. CAD = coronary artery disease. ST = ST segment elevation / depression (mV). Significance levei between study groups: ^* ~ p < 0.05.

Figure 18: Results of the individual linear regression analyses of the morphological descriptors to RR-interval. Difference between exercise and recovery * p < 0.05; ^** p < 0.01.

Figure 19: Correlation coefficients between electrocardiographic measurements during exercise. Significance of correlation ^* p < 0.05 and ^** p < 0.01.

Figure 20: Correlation coefficients between electrocardiographic measurements during recovery. Significance of correlation ^* p < 0.05 and ^** p < 0.01. It will be obvious to a person skilled in the art that, as technology advances, the inventive concept can be implemented in various ways. The invention and its embodiments are not iimited to the examples described above but may vary within the scope of the claims.

Claims

Claims

1. An apparatus comprising a processor configured to calculate morphological parameter values of QRS complex and T wave from a sample period within electrocardiography data recorded from a subject; to calculate heart rate values from the sample period; to form an association result between the change of the morphological parameter values and change of the heart rate values; and to predict a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition, wherein the morphological parameter values comprise at least one of the following: vaiues of total cosine R-to-T, values of QRS complex loop asymmetry, values of T wave loop dispersion, values of QRS complex and T wave principal component analysis, values of an angle between a main QRS complex loop vector and a main T wave loop vector,

2. The apparatus of claim 1 , wherein the sampie period comprises a recovery period of the subject, and/or sub-sample periods within the electrocardiography data.

3. The apparatus of any preceding claim, wherein the processor is further configured to form the comparison result by forming a correlation value between the change of the morphological parameter values and change of the heart rate values.

4. The apparatus of claim 3, wherein the processor is further configured to analyze the way the comparison result fulfils the predetermined condition by comparing the correlation value to a predetermined correlation threshold value.

5. The apparatus of any preceding claim, wherein the morphological parameter values of QRS complex and T wave are calculated separately from two sample periods, from an exercise period and from a recovery period, within electrocardiography data recorded from a subject, heart rate values are calculated from the exercise period and from the recovery period, association results are calculated between the change of the morphoiogical parameter values and change of the heart rate values for the exercise period and for the recovery period, a combined association result is formed from the association result calculated for the exercise period and from the association result calculated for the recovery period, and the risk to the cardiac death of the subject is predicted on the basis of the way the combined association result fulfils a predetermined condition for the combined association result.

6. A method comprising: calculating morphological parameter values of QRS complex and T wave from a sample period within electrocardiography data recorded from a subject; calculating heart rate values from the sample period; forming an association result between the change of the morphological parameter values and change of the heart rate values; and predicting a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition, wherein the morphological parameter values comprise at least one of the following: values of total cosine R-to-T, values of QRS complex loop asymmetry, values of T wave loop dispersion, values of QRS complex and T wave principal component analysis, values of an angle between a main QRS complex loop vector and a main T wave loop vector.

7. The method of claim 6, wherein the sample period comprises a recovery period of the subject, and/or sub-sample periods within the electrocardiography data.

8. The method of claim 6 or 7, further comprising: forming the comparison result by forming a correlation value between the change of the morphological parameter values and change of the heart rate values.

9. The method of claim 8, further comprising: analyzing the way the comparison result fulfils the predetermined condition by comparing the correlation value to a predetermined correlation threshold value.

10. The method of any preceding claim 6 to 9, wherein the morphoiogical parameter values of QRS complex and T wave are calculated separately from two sample periods, from an exercise period and from a recovery period, within electrocardiography data recorded from a subject, heart rate values are calculated from the exercise period and from the recovery period, association results are calculated between the change of the morphological parameter values and change of the heart rate values for the exercise period and for the recovery period, a combined association result is formed from the association result calculated for the exercise period and from the association result calculated for the recovery period, and the risk to the cardiac death of the subject is predicted on the basis of the way the combined association result fulfils a predetermined condition for the combined association result.

11. A computer program comprising program instructions which, when loaded into an apparatus, cause the apparatus to perform the process of any preceding claim 6 to 10.

12. A computer program on a carrier, comprising program instructions which, when loaded into an apparatus, cause the apparatus to calculate morphological parameter values of QRS complex and T wave from a sample period within electrocardiography data recorded from a subject; to calculate heart rate values from the sample period; to form an association result between the change of the morphological parameter values and change of the heart rate values; and to predict a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition, wherein the morphological parameter values comprise at least one of the following: values of total cosine R-to-T, values of QRS complex loop asymmetry, values of T wave loop dispersion, values of QRS complex and T wave principal component analysis, values of an angle between a main QRS complex loop vector and a main T wave loop vector.

13. An apparatus comprising: means for calculating morphological parameter values of QRS complex and T wave from a sample period within electrocardiography data recorded from a subject; means for caiculating heart rate values from the sample period; means for forming an association result between the change of the morphological parameter values and change of the heart rate values; and means for predicting a risk to a cardiac death of the subject on the basis of the way the association result fulfils a predetermined condition, wherein the morphological parameter values comprise at least one of the following: values of total cosine R-to-T, values of QRS complex loop asymmetry, values of T wave loop dispersion, values of QRS complex and T wave principal component analysts, values of an angie between a main QRS complex loop vector and a main T wave loop vector.