Methods and systems for myocardial infarction prediction

Abstract

A signal is received 302 from a non-invasive heart sensor, positioned to detect the heartbeat of a driver. It is used to predict the likelihood of myocardial infarction, inferring likelihood of a road traffic accident. A plurality of values for heart rate n-variability (HRnV) 304 and average heart rate n-variability (avHRnV) 306 are determined, from RR intervals derived from the heart rate signal. The intervals have summation parameter n and stride parameter m, and the plurality of values includes different values of at least one of n and m to allow for consideration of HRV behaviour in varying scales. The values are passed to a predictive model 308, resulting in an alert if the likelihood exceeds a threshold. The heart rate sensor may be contactless, e.g. an infrared detector, and configured for use in a motor vehicle. The predictive model may also take medical history and demographic information into account.

Description

Technical Field

[0001] The present invention relates, in general terms, to systems and methods for monitoring of cardiovascular health and conditions of a driver of a motor vehicle. In particular, the present invention relates to monitoring and prediction of myocardial infarction (Ml) or heart attack of a driver of a motor vehicle in real-time. Background

[0002] This background is provided for generally presenting the context of the disclosure. Contents of this background section are neither expressly nor implied admitted as prior art against the present disclosure.

[0003] Motor vehicle accidents, often referred to as car accidents or traffic collisions can result in property damage, injuries and even fatalities. Road safety and driver safety are key considerations when it comes to automotive industry and public safety. Specifically, monitoring health-related conditions, especially acute cardiac events, of the driver is of great interest, since conditions, such as acute heart attacks, present a critical challenge to public safety on roads. Traditionally, there were extremely limited means to monitor the health conditions of the driver accurately, timely, and pragmatically. However, with the rise of smart sensors and artificial intelligence, it is technically and economically viable to monitor the driver’s health in real-time.

[0004] The current available technology systems and methods lack valid methods to detect or predict adverse health conditions from physiological data captured by real-time monitoring systems. There are currently no functional solutions for driver health monitoring. Current clinical solutions for Ml prediction require invasive tests and accurate electrocardiogram (ECG) recordings, rendering them unsuitable for use in in automotives.

[0005] It would be desirable to overcome or ameliorate at least one of the above-described problems, or at least to provide a useful alternative. Summary

[0006] Disclosed is a system for predicting a likelihood of a road traffic accident, comprising: memory; at least one processor (processor(s)); and a prediction module comprising a predictive model trained to predict a likelihood of a road traffic accident, wherein the memory stores instructions that, when executed by the processor(s), cause the processor(s) to: receive a heart rate signal, comprising a plurality of heart beats, from a non-invasive heart rate sensor, the non-invasive heart rate sensor being position to detect a heart beat of a driver; determine a value for each of a plurality of heart rate n-variability (HRnVm) and average heart rate n-variability (avHRnV) parameters from the heart rate signal, for a plurality of values of at least one of n and m, where n and m are natural numbers and m <n, apply the predictive model to the values of the HRnVm and avHRnV parameters, to predict a likelihood of road traffic accident due to myocardial infarction occurrence for the driver; and output a signal alerting the driver if the predicted likelihood exceeds a predetermined threshold.

[0007] Disclosed is a system for predicting a likelihood of myocardial infarction, comprising: memory; at least one processor (processor(s)); and a prediction module comprising a predictive model trained to predict a likelihood of myocardial infarction, wherein the memory stores instructions that, when executed by the processor(s), cause the processor(s) to: receive the heart rate signal, comprising a plurality of heart beats, from a non-invasive heart rate sensor, the non-invasive heart rate sensor being position to detect a heart beat of a driver; determine a value for each of a plurality of heart rate n-variability (HRnVm) and average heart rate n-variability (avHRnV) parameters from the heart rate signal, for a plurality of values of at least one of n and m, where n and m are natural numbers and m <n; apply the predictive model to the values of the HRnVm and avHRnV parameters, to predict a likelihood of myocardial infarction occurrence for the driver; and output a signal alerting the driver if the predicted likelihood exceeds a predetermined threshold.

[0008] Disclosed is a method for predicting a likelihood of myocardial infarction, the method comprising: receiving a heart rate signal, detected by a non-invasive heart rate sensor, the signal comprising a plurality of heart beats of a driver; determining a value for each of a plurality of heart rate n-variability (HRnVm) and average heart rate n-variability (avHRnV) parameters from the heart rate signal, for a plurality of values of at least one of n and m, where n and m are natural numbers and m <n; applying the predictive model to the values of the HRnVm and avHRnV parameters, to predict a likelihood of myocardial infarction occurrence for the driver, the predictive model being trained to predict a likelihood of myocardial infarction; and output a signal alerting the driver if the predicted likelihood exceeds a predetermined threshold.

[0009] Disclosed is a method for predicting a likelihood of road traffic accident, the method comprising: receiving a heart rate signal, detected by a non-invasive heart rate sensor, the signal comprising a plurality of heart beats of a driver; determining a value for each of a plurality of heart rate n-variability (HRnVm) and average heart rate n-variability (avHRnV) parameters from the heart rate signal, for a plurality of values of at least one of n and m, where n and m are natural numbers and m <n; applying the predictive model to the values of the HRnVm and avHRnV parameters, to predict a likelihood of road traffic accident due to myocardial infarction occurrence for the driver; and output a signal alerting the driver if the predicted likelihood exceeds a predetermined threshold. Brief description of the drawings

[0010] Some embodiments of systems and methods for myocardial infarction prediction, in accordance with the present disclosure, will now be described, by way of non-limiting example, with reference to the drawings in which:

[0011] Figure 1 illustrates a demonstration of the generation of RR3I2 or avRR3I2 from a RRI series;

[0012] Figure 2 illustrates an overview of variable preparation and selection process;

[0013] Figure 3 is a flow diagram of a method according to an embodiment of the invention;

[0014] Figure 4 illustrates receiver operating characteristic (ROC) curves of the models with and without avHRnV extension;

[0015] Figure 5 illustrates a system for implementing the methods of Figure 1; and

[0016] Figure 6 illustrates a method for training a predictive model used by the system of Figure 5 in predicting a likelihood of road traffic accident due to Ml, or a likelihood of Ml itself. Detailed description

[0017] Embodiments relate to systems and methods for monitoring of cardiovascular conditions, especially in predicting acute cardiac events such as myocardial infarction (Ml) among drivers of motor vehicles, with the intention to alert drivers if the likelihood of road traffic accident is high (i.e., above a predetermined threshold), due to a correspondingly high likelihood of Ml. Embodiments relate to systems and methods suitable for real-time implementation in automotives. The ability to predict such acute cardiac events among drivers in real-time by processing data obtained non-invasively while driving motor vehicles can allow for appropriate actions to be taken by drivers to pull over to a safe place, or get immediate medical attention, and consequently avoid motor vehicle accidents. The system and methods may operate in real-time, for practically immediate response to likely onset of Ml.

[0018] Hereinafter, the systems and methods for real-time prediction of the likelihood of road traffic accident, or Ml, for a driver of a motor vehicle will be described in detail with reference to Figure 1 to Figure 5 according to the preferred embodiments. It is to be understood that limiting the description to the preferred embodiments of the invention is merely to facilitate discussion of the present invention. Various modifications may be made, to the preferred embodiments, without departing from the scope of the appended claims.

[0019] Clinically, likelihood or risk of Ml is usually predicted using patients’ vital signs, medical history, blood tests, and other clinical tests. Heart rate variability (HRV) has been found to be useful in improving Ml prediction, and thus in predicting the likelihood of accidents resulting from Ml. HRV evaluates changes in cardiac autonomic regulation, and is closely related to the autonomic nervous system (ANS). HRV studies the variation of heartbeat intervals or R-to-R peak intervals (RRIs) and, in a clinical setting, would be derived from ECG signals. However, ECG machines are generally unsuitable for use in cars and other vehicles or automotives. For this reason, the heartbeat sensors referred to in the present disclosure are non-invasive sensors that acquire heartbeat signals through non-invasive means - e.g., a smartwatch with heart rate detector, or an infrared sensor. The heart rate sensors referred to herein may also therefore be contactless sensors - e.g., infrared detector that does not require skin contact. It is believed that a decrease of variations in HRV is usually correlated with morbidity and mortality. In a preferred embodiment, PhysioNet Cardiovascular Signal Toolbox is used for RRI preprocessing and conventional HRV metrics calculations.

[0020] Embodiments incorporate an artificial intelligence (Al) module, average heart rate n-variability (avHRnV) metrics to predict acute cardiac events, and thus, in some embodiments, to infer a likelihood of road traffic accident resulting from Ml. This Al module may also take into account heart rate n-variability (HRnV) metrics (to complement HRV analysis by considering HRV behaviors in varying scales), medical history, demographic information, patient preliminary physiological measurements and others. The physiological meanings and the processing mechanism of HRnV are not clear in every detail. One of the disadvantages of using HRnV is its use of summation windows, which increases some metrics in time and frequency domains as the parameters of HRnV increase, making the interpretation and comparison between different groups of HRnV metrics difficult. Herein, avHRnV is introduced as an extension of HRnV to address this issue and further improve the performance of HRnV with its application in Ml prediction for driver’s health monitoring.

[0021] In some embodiments, a list of variables or metrics is determined and input into a logistic regression algorithm to build the predictive model - e.g., a logistic model. The logistic regression algorithm may be a stepwise logistic regression algorithm, and / or may be multivariate. In other embodiments, the predictive model may apply kernel methods to HRnV and avHRnV metrics to capture the nonlinear interactions introduced by avHRnV and to improve the accuracy of the prediction model without the need of acquiring additional measurements from drivers.

[0022] The predictive model may adopt kernel functions (e.g. polynomial kernel and radial basis function kernel) to combine HRnV and avHRnV parameters in a nonlinear manner. The rationale for this algorithm is that some HRnV and avHRnV parameters are linearly linked, which could reduce the discriminatory power in predicting adverse clinical outcomes. With HRnV, avHRnV, and avHRnVm as inputs, various types of kernel function-mapped new nonlinear feature vectors can be created, e.g., k(HRnV, avHRnV), k(HRnV, avHRnVm), where k() is the kernel function and HRnV, avHRnV and avHRnVm are vectors of identical dimensions. Notably, HRnV for n = 1 is HRV and, similarly, avHRnVm for n = m = 1 is HRV. Thus, with appropriate parametrization, the comparisons can include relationships across tuples involving HRV, HRnV, HRnVm, avHRnV and avHRnVm, though pairs will be used for illustration purposes.

[0023] Multiple kernel functions may be used such that multiple sets of converted feature vectors are available. When choosing kt as the kernel function, the parameters / metrics may include fc(HRnV, avHRnV), kf(HRnV, avHRnVm), where "t" is an integer of any value depending on the choice of kernel functions. Newly created kernel features may be integrated with basic patient information and clinical investigation results as inputs to feed into a prediction algorithm, such as logistic regression (particularly where stepwise logistic regression has been used for selection of variables in the model), neural network, support vector machine, decision tree, random forest, boosting, or ensemble learner etc.

[0024] Each HRnV or avHRnV analysis may be a conventional HRV analysis performed on the corresponding interval sequence (i.e., RRI or RRnIm), some embodiments group the metrics from each HRnV or avHRnV analysis as a metrics vector. For a single patient, multiple HRnV and multiple avHRnV analyses can be performed using the same source RRI measured from the patient, resulting in a number of metrics vectors (vector representations) of the same length (e.g., HRnV metrics vector, HR3V2 metrics vector). Kernel methods can then be applied on these metrics vectors to reflect the nonlinear interactions between them.

[0025] Considering two different HRnV and / or avHRnV metrics vectors obtained from the same patient (V^avyHR Vk and V^avyHRnVm will be used for illustration purposes) of the same length, kernel methods involve calculating the inner product of the two vectors in a high dimensional vector space, measuring the nonlinear interactions between the vectors in their original space. The inner product I^xnm from kernel kt is given by: Ijkxnm =

[0026] Notably, the present methods may use a pair of HRnV metrics, or a pair of anHRnV metrics, or may use both HRnV and avHRnV metrics.

[0027] Some embodiments use one or more than one of six different forms of kernel forfct. The six different forms include cosine similarity, polynomial kernel, sigmoid kernel, RBF kernel, Laplacian kernel, and Chi-squared kernel. The inner products obtained from these kernel methods can then be treated as candidates (inputs / parameters) for the stepwise model or the prediction model.

[0028] In some embodiments, HRnV and avHRnV analyses were performed on the source RRI from each of the patients with both HRnV and avHRnV parameters n and m up to 3, resulting in 6 different HRnV metrics vectors (including the conventional HRV metrics vector), and 5 avHRnV vectors. The nonlinear interactions were measured between every unique pair from the 11 metrics vectors, providing 55 inner products for each of the kernel methods. In total, 330 different inner products were added to each patient’s data.

[0029] Figure 6 provides a graphical illustration of various steps of kernel methods. At step 610, R-R interval signals (heart beat data or signals) is obtained from a non-invasive sensor. The received data is subjected to cleaning and signal processing operations at step 612, if necessary. These signal processing steps can be standard filtration and noise removal steps and others. Step 614, HRnVm (includes HRV and HRnV, where n=1 and / or m=1) and avHRnVm (includes avHRnVm, where m = 1) metrics are computed. At step 616, the HRnV and avHRnV metrics data is transformed into vectors and subjected to kemelization using any one or more of the method identified above. Examples of kernel metrics are illustrated in blocks 616(1) and 616(t). One or more kernel metrics may be computed based on various potential combinations of HRnV and avHRnV metrics. The kernel metrics may be combined with any driver (patient) data, clinical information and other vital signs, symptoms and relevant information, and provided input to the prediction module at step 620 to determine a likelihood (also referred to as "risk") of road traffic accident due to Ml, or Ml itself, for the driver. If the Ml risk is above a predetermined threshold - e.g., where the predictive model outputs a likelihood of Ml between 0 and 1, the predetermined threshold may be 0.2 or 0.5 or any other desired threshold which may, for example, be experimentally determined - the predicted likelihood is outputted at 622 in the form of an alert. The same threshold determination can be used to instead instruct the driver to pull to the side of the road, to a clinic, or other area where the car is safely located and, ideally, the driver's condition can be treated. The alert may be produced by pushing an alarm or notification to the driver's mobile device, automatically alerting a clinic or physician (e.g. by pushing a notification to the driver's mobile device, that then triggers an application on the mobile device to send the alert to a predetermined mobile device number or in the form of an email to the email address of an emergency department).

[0030] In a study of the present methods, a retrospective analysis was conducted of data collected from a sample of patients presented to the Emergency Department (ED) of Singapore General Hospital (SGH) from September 2010 to July 2015. SGH is the largest tertiary care hospital in Singapore, with an ED that sees between 300-500 patients daily. All patients recruited in the study were classified using the Patient Acuity Category Scale (PACS), with PACS 1 patients being the most critically ill and PACS 4 patients being non-urgent. The 795 patients finally included in this study were all classified to be PACS 1 or 2. The patients’ symptoms description, medical history (including cardiac disease history), vital signs taken at the ED, and RRI data derived from electrocardiogram (ECG) recorded at ED were included in this study. The decision to include these variables was made to mimic the data acquisition capability in an out-of-hospital setting, similar to the setting in automotives, where no blood test or ECG recording can be performed. Table 1 provides an exhaustive overview of the data included in this study. The primary endpoint for the prediction model was Ml within 30 days (30-day Ml) from the patient’s presentation. Variable Category Variable Name Description Value type Demographics age Age of the patient Integer; numerical gender Gender of the patient Male or female; categorical smoking If the patient is an active smoker, and how long the Categorical patient has been smoking Symptoms symp_cp_peak_onset_24h Chest pain within 24 hours of presentation Yes or no; categorical symp_cp_radiating_arm_neck Chest pain radiating to left arm or neck Yes or no; categorical symp_cp_rest Chest pain occurring at rest Yes or no; categorical symp_shortness_of_breath Shortness of breath Yes or no; categorical symp_dizziness Having dizziness Yes or no; categorical symp_nausea Feeling nausea Yes or no; categorical symp_palpitations Having palpitations Yes or no; categorical Vital Signs vitals_temp Temperature reading Numerical vitals pulse bpm Pulse in bpm Numerical vitals rr Respiratory rate Numerical vitals_sbp Systolic blood pressure Numerical vitals_dbp Diastolic blood pressure Numerical vitals spo2 SpO2 Numerical vitals_o2_supplementation 02 supplementation in percentage Numerical vitals_pain_score Pain Score Numerical Medical History medhx_diabetes History of diabetes Yes or no; categorical medhx_hypertension History of hypertension Yes or no; categorical medhx_dyslipidaemia History of dyslipidemia Yes or no; categorical medhx_stroke History of stroke Yes or no; categorical medhx_cancer History of cancer Yes or no; categorical medhx_respiratory_disease History of respiratory disease Yes or no; categorical medhx_renal_disease History of renal disease Yes or no; categorical heart_history History of ACS or ACS related incidents Yes or no; categorical RRI rri 5 or 6 minutes of RR intervals converted from ECG recordings acquired at ED Numerical sequence Table 1 -- Descriptions of data The HRnV Method and the avHRnV Extension

[0031] Embodiments comprise processing of the heartbeat signals of drivers obtained from infrared sensors or any other contactless sensors to convert the original heartbeat signals to R-R interval (RRI) sequences.

[0032] The HRnV method for alternative RRI representation for HRV utilizes sliding and stridden summation windows on the original RRI. The resulted intervals are denoted as RRnIm intervals, which can be fed into conventional HRV analysis as normal RRI to obtain HRnVm metrics. For clarification, the term ‘HRnV’ refers to the name of the method (i.e., heart rate n-variability) and the metrics calculated using this method, while HR„Vm refer to the derived metrics based on the / ? / ?„ / m intervals with specified n and m.

[0033] The summation window used in the original HRnV generates RRnIm with increasing amplitude as n increases, making comparison between different groups of HRnV metrics (e.g., comparison between HR2V1 and HR^ metrics) difficult. In a preferred embodiment, extension of HRnV is introduced to use averaging window instead of summation. The resulting intervals and metrics are denoted as avRRnIm and avHRnVm respectively.

[0034] The HRnV method and the avHRnV extension require specification of two parameters: the summation parameter n and the stride parameter m. Bothn and m can take any positive integer values (i.e., n,m >= 1) though m must be less than n or less than or equal to n (m<=n). To describe the RRnIm intervals generation used in HRnV and avHRnV, consider a series of clean input RRI as Xt (i = 1,2,3,-, N) of length N. With specified parameters n and m, a new series of RRnIm or avRRnIm intervals, Yt (i = 1,2,3,...,M) of length M (M <= N) can be expressed as: (£_1>m+7, (1 = 1,2,3,

[0035] The weight of each data point, w, is set to 1 when using the original HRnV method and ± when using the avHRnV method. The length of the new intervals, M, is given by M = , where [ • J represents the floor function. When the parameters n and m are equal, the RRnIm (or / ? / ?n / n)intervals can be abbreviated as RRnI. Consequently, metrics calculated based on RRnl can be abbreviated as HRnV. Essentially, the original RRI is equivalent to RRJ intervals, and HRV metrics are equivalent to HR±V metrics. Figure 1 presents a demonstration of the generation of RR3I2 or avRR3I2 from a toy RRI series.

[0036] The RRnIm and avRRnIm intervals can be analyzed as normal RRIs using the conventional HRV analysis process. Metrics characterizing these intervals can be categorized in three different domains, and the detailed list of such metrics can be found in Table 2 as shown below. All the processing and metrics calculation of RRI are done using the HRnV-Calc software with modified functionalities for avHRnV. All HRnV and avHRnV metrics used in this study were calculated with all combinations of parameters n and m being not greater than 3. Category Metrics Name Description Units Time Average RR The mean of RRnIm intervals ms Domain SDRR The standard deviation of RRnIm intervals ms Average HR The mean of heart rates 1 / min SDHR The standard deviation of heart rates 1 / min RMSSD Square root of the mean squared differences between successive RR intervals ms NN50(x) Numbers of RRnIm intervals differ more than 50 ms from the previous intervals. ForNN50x, the difference will be set to x times of 50 ms, where x = n in the corresponding HRnVm analysis count pNN50(x) Percentage of NN50(x) intervals within the entire RRnIm intervals % RR Skewness The skewness of the RRnIm intervals distribution - RR Kurtosis The kurtosis of the RRnIm intervals distribution - RR Triangular Index The integral of the RRnIm intervals histogram divided by the height of the histogram - Frequency Domain VLF, LF, and HF Peak frequencies The peak frequencies in the power spectral distribution (PSD) for VLF, LF, and HF bands Hz VLF, LF, and HF Powers Absolute powers of VLF, LF, and HF bands ms^ VLF, LF, and HF Power Percentages The percentage for powers of VLF, LF, and HF bands within the overall spectrum % LF and HF Normalized Powers Normalized powers for LF and HF bands n.u. Total Power The overall power of the PSD ms2 lf / hf The ratio between the powers of LF and HF bands Nonlinear Domain Poincare SD1 and SD2 The width and length of the eclipse fitted in the Poincare plot ms ApEn Approximate entropy - SampEn Sample entropy - DFA and «2 Short-term and long-term fluctuations of detrended fluctuation analysis (DFA) - Table 2 -- Descriptions of HRV, HRnV, and avHRnV Metrics Statistical analysis

[0037] Figure 2 provides a comprehensive overview 200 of the variable preparation and selection process. All variables selected for this study as shown in Table 1 (202), including HRV, HRnV, and avHRnV metrics (208) calculated from the original RRI (206) acquired from patients (i.e., drivers), were first analyzed using univariate tests (204). Each variable was evaluated as an individual predictor of the primary outcome (30-day Ml). For categorical variables, the Chi-Square test was used, while the Mann-Whitney U-test was used for numerical variables. Only the variables with p-value no greater than 0.4 were selected for the subsequent analysis, where collinearity within the variables were eliminated based on Pearson’s correlation coefficients calculated for every unique pair of the selected variables (210). For a pair of variables with Pearson’s correlation coefficient being no less than 0.85, the variable with higher p-value was eliminated.

[0038] The final list of variables was then input into multivariate stepwise logistic regression with backward elimination (212) to build the predictive model. Backward elimination was chosen for the stepwise variable selection because it has the advantage to assess the joint predictive ability of variables, and it removes the least essential variables. The eliminated variables were not allowed to re-enter the model in some embodiments. In other embodiments, every combination of variables was examined. Examination of every combination of variables required significant computing resources with the risk of overfitting the model when the number of variables is large. The goal for the prediction model was Ml within 30 days from the patient’s ordriver's presentation (214).

[0039] Figure 3 illustrates a flow diagram of a method for predicting myocardial infarction according to an embodiment of the invention. The method comprises receiving a heart rate signal from a heart rate sensor. The signal includes a plurality of heart beats from a user or a motor vehicle driver (step 302). A value for each of a plurality of heart rate variability (HRV) parameters is then determined, which can be or include a plurality of HRnV parameters (step 304) and a plurality of avHRnV parameters (step 306). Steps 304 and 306 can occur concurrently, since all relevant metrics can be derived from the original heart beat signal. The predictive model (e.g., a model trained as set out above, which may be a logistic regression model) is then applied to the values to predict a likelihood of myocardial infarction occurrence for the user (step 308). The likelihood, if above a predetermined threshold, will then be outputted in the form of an alarm or alert (step 310). Step 310, in some embodiments, can be replaced with outputting an alert to the driver, road traffic authority and / or clinic, advising the driver to move to a safe location for the car. In these embodiments, determining the likelihood of Ml may be sufficient to then infer that a road traffic accident is likely and thus trigger the alert. Results

[0040] To examine the usefulness of adding avHRnV metrics to the HRnV metrics family, two groups of variables are prepared: with and without avHRnV metrics. The group with avHRnV metrics contained 365 variables with 199 of them having p-values no more than 0.4. After collinearity elimination, 78 variables were kept for stepwise regression, which selected 21 variables for the final predictive model. The groups without avHRnV metrics contained 210 variables with 125 of them having p-values no more than 0.4. 68 variables were kept after collinearity elimination, and 21 of them were selected for the final model. Table 3 contains the full lists of variables selected in the predictive models for both groups. The two predictive models were compared using 10-fold cross-validation.

[0041] Figure 4 shows the ROC curves of the two models. The average area under the curve (AUC) performance of the model with avHRnV metrics is 0.770, while the average performance of the model without avHRnV is 0.739. Model without avHRnV Extension OR (95% Cl) P-value Adjusted_OR (95% ci) Adjusted p-value age 1.012(0.999-1.026) 0.074 1.041 (1.021-1.061) 0 gender 1.737 (1.171-2.576) 0.006 1.545 (0.922-2.589) 0.098 smoking 1.181 (1.039-1.343) 0.011 1.174 (0.996-1.384) 0.056 sympdizziness 0.805(0.522-1.241) 0.325 0.632 (0.377-1.059) 0.081 symp_nausea 1.564 (1.003-2.441) 0.049 1.935 (1.119-3.347) 0.018 vitals_sbp 1.004(0.998-1.01) 0.163 0.986 (0.975-0.997) 0.015 vitals_dbp 1.021 (1.009-1.033) 0 1.043 (1.02-1.067) 0 vitals_o2_supplementation 1.02 (1.002-1.038) 0.025 1.025 (1.002-1.048) 0.035 vitals_pain_score 1.086 (1.024-1.152) 0.006 1.11 (1.035-1.19) 0.003 heart_history 9.466 (4.11-21.801) 0 13.681 (5.49-34.097) 0 medhx_diabetes 1.494 (1.056-2.115) 0.023 1.691 (1.096-2.608) 0.018 medhxdyslipidaemia 0.782 (0.555-1.102) 0.16 0.631 (0.416-0.956) 0.03 medhx_cancer 0.429 (0.128-1.438) 0.17 0.25 (0.066-0.945) 0.041 medhx_respiratory_disease 0.381 (0.114-1.269) 0.116 0.165 (0.041-0.671) 0.012 h rvrrs kewness 1.101 (1.031-1.176) 0.004 1.124 (1.029-1.228) 0.01 hr2v_sam_ent 1.0 (1.0-1.0) 0.193 1.0 (1.0-1.0) 0.018 hr2v1_lf_hf_ratio 0.945 (0.881-1.014) 0.116 1.221 (1.026-1.453) 0.025 hr2v1_app_ent 2.533 (0.837-7.667) 0.1 5.665 (1.417-22.643) 0.014 hr2v1_dfa_a1 0.714(0.384-1.327) 0.287 5.023 (1.648-15.312) 0.005 hr3v2_avhr 1.055 (1.022-1.088) 0.001 1.1 (1.052-1.151) 0 hr3v2_lf_hf_ratio 0.914 (0.858-0.974) 0.005 0.679 (0.582-0.793) 0 Model with avHRnV Extension age 1.012(0.999-1.026) 0.074 1.042 (1.023-1.062) 0 smoking 1.181 (1.039-1.343) 0.011 1.174 (1.003-1.376) 0.046 symp_palpitations 0.599 (0.331-1.086) 0.092 0.531 (0.267-1.058) 0.072 vitals_sbp 1.004 (0.998-1.01) 0.163 0.985 (0.973-0.996) 0.007 vitals_dbp 1.021 (1.009-1.033) 0 1.047 (1.024-1.071) 0 vitals_o2_supplementation 1.02 (1.002-1.038) 0.025 1.021 (0.999-1.044) 0.067 vitals_pain_score 1.086 (1.024-1.152) 0.006 1.113 (1.036-1.196) 0.003 heart_history 9.466 (4.11-21.801) 0 14.077 (5.584- 35.487) 0 medhx_diabetes 1.494 (1.056-2.115) 0.023 1.694 (1.087-2.641) 0.02 medhxdyslipidaemia 0.782(0.555-1.102) 0.16 0.649 (0.425-0.99) 0.045 medhx_cancer 0.429(0.128-1.438) 0.17 0.234 (0.062-0.883) 0.032 medhx_respiratory_disease 0.381 (0.114-1.269) 0.116 0.218 (0.053-0.899) 0.035 hrv_rr_skewness 1.101 (1.031-1.176) 0.004 1.123 (1.027-1.229) 0.011 hr2v1_lf_hf_ratio 0.945 (0.881-1.014) 0.116 1.239 (1.039-1.477) 0.017 hr3v2_lf_hf_ratio 0.914(0.858-0.974) 0.005 0.672 (0.575-0.786) 0 avhr2v_sam_ent 1.0 (1.0-1.0) 0.193 1.0 (1.0-1.0) 0.009 avhr2v1_app_ent 2.533 (0.837-7.667) 0.1 4.693 (1.161-18.971) 0.03 avhr2v1_dfa_a1 0.714(0.384-1.327) 0.287 5.099 (1.612-16.127) 0.006 avhr3v2_avhr 1.018(1.007-1.029) 0.001 1.037 (1.021-1.053) 0 avhr3v2_lf_peak 0.0 (0.0-0.0) 0 0.0 (0.0-0.0) 0 avhr3v2_lf_per 1.009 (0.996-1.021) 0.184 1.016 (1.001-1.032) 0.037 Table 3 -- Final variable lists for the predictive models and corresponding odds ratios (OR) and p-values

[0042] Figure 5 illustrates a system for implementing the method 300. The system 500 comprises a heart rate sensor 510, for detecting the heart rate from a driver and producing a signal 520 based on the heart rate. The signal is received by a system 530 for predicting a likelihood of Ml of the driver from whom the signal was acquired - in some embodiments, the system 530 may include the sensor 510. The signal is processed by the processor(s) 532, using program code 538 stored in memory 534, to determine a plurality of values corresponding to HRnV and avHRnV parameters. A predictive model 536, also stored in memory 534, processes the values to determine a likelihood of Ml of the driver based on the heart rate signal, or of a road traffic accident resulting from Ml. If the likelihood of Ml exceeds a predetermined threshold, an alert is outputted in the form of an alarm as discussed above, by alert system 540.

[0043] It will be appreciated that many further modifications and permutations of various aspects of the described embodiments are possible. Accordingly, the described aspects are intended to embrace all such alterations, modifications, and variations that fall within the spirit and scope of the appended claims.

[0044] The reference in this specification to any prior publication (or information derived from it), or to any matter which is known, is not, and should not be taken as an acknowledgment or admission or any form of suggestion that that prior publication (or information derived from it) or known matter forms part of the common general knowledge in the field of endeavour to which this specification relates.

[0045] Throughout this specification and the claims which follow, unless the context requires otherwise, the word “comprise”, and variations such as “comprises” and “comprising”, will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

[0046] The reference in this specification to any prior publication (or information derived from it), or to any matter which is known, is not, and should not be taken as an acknowledgment or admission or any form of suggestion that that prior publication (or information derived from it) or known matter forms part of the 5 common general knowledge in the field of endeavour to which this specification relates.

Claims

1. A system for predicting a likelihood of a road traffic accident, comprising:memory;at least one processor (processor(s)); anda prediction module comprising a predictive model trained to predict a likelihood of a road traffic accident,wherein the memory stores instructions that, when executed by the processor(s), cause the processor(s) to:receive a heart rate signal, comprising a plurality of heart beats, from a non-invasive heart rate sensor, the non-invasive heart rate sensor being position to detect a heart beat of a driver;determine a value for each of a plurality of heart rate n-variability (HRnVm) and average heart rate n-variability (avHRnV) parameters from the heart rate signal, for a plurality of values of at least one of n and m, where n and m are natural numbers and m <n;apply the predictive model to the values of the HRnVm and avHRnV parameters, to predict a likelihood of road traffic accident due to myocardial infarction occurrence for the driver; andoutput a signal alerting the driver if the predicted likelihood exceeds a predetermined threshold.

2. A system for predicting a likelihood of myocardial infarction, comprising:memory;at least one processor (processor(s)); anda prediction module comprising a predictive model trained to predict a likelihood of myocardial infarction,wherein the memory stores instructions that, when executed by the processor(s), cause the processor(s) to:receive the heart rate signal, comprising a plurality of heart beats, from a non-invasive heart rate sensor, the non-invasive heart rate sensor being position to detect a heart beat of a driver;determine a value for each of a plurality of heart rate n-variability (HRnVm) and average heart rate n-variability (avHRnV) parameters from the heart rate signal, for a plurality of values of at least one of n and m, where n and m are natural numbers and m <n:apply the predictive model to the values of the HRnVm and avHRnV parameters, to predict a likelihood of myocardial infarction occurrence for the driver; andoutput a signal alerting the driver if the predicted likelihood exceeds a predetermined threshold.

3. The system of claim 1 or 2, wherein the predictive model is trained to predictthe likelihood from a training dataset comprising at least one of patientdemographic data and clinical data relating to myocardial infarction.

4. The system of claim 3, wherein the processor(s) extract the training datasetfrom past data of a pool of patients, and trains the predictive model to identifypatterns in the values corresponding to myocardial infarction occurrence reflected in the past data.

5. The system of any one of claims 1 to 4, wherein the heart rate sensor is contactless.

6. The system of any one of claims 1 to 5, wherein the heart rate sensor is configured for use in a motor vehicle to obtain the heart rate signal.

7. A method for predicting a likelihood of myocardial infarction, the method comprising:receiving a heart rate signal, detected by a non-invasive heart rate sensor, the signal comprising a plurality of heart beats of a driver;determining a value for each of a plurality of heart rate n-variability (HR„Vm) and average heart rate n-variability (avHRnV) parameters from the heart rate signal, for a plurality of values of at least one of n and m, where n and m are natural numbers and m <it,applying the predictive model to the values of the HRnVm and avHRnV parameters, to predict a likelihood of myocardial infarction occurrence for the driver, the predictive model being trained to predict a likelihood of myocardial infarction; andoutput a signal alerting the driver if the predicted likelihood exceeds a predetermined threshold.

8. A method for predicting a likelihood of road traffic accident, the method comprising:receiving a heart rate signal, detected by a non-invasive heart rate sensor, the signal comprising a plurality of heart beats of a driver;determining a value for each of a plurality of heart rate n-variability (HRnVm) and average heart rate n-variability (avHRnV) parameters from the heart rate signal, for a plurality of values of at least one of n and m, where n and m are natural numbers and m <it,applying the predictive model to the values of the HRnVm and avHRnV parameters, to predict a likelihood of road traffic accident due to myocardial infarction occurrence for the driver; andoutput a signal alerting the driver if the predicted likelihood exceeds a 5 predetermined threshold.

9. The method of claim 7 or 8, further comprising training the predictive model to predict the likelihood, using a training dataset comprising at least one of patient demographic data and clinical data relating to myocardial infarction.

10. The method of claim 9, further comprising extracting the training dataset from 10 past data of a pool of patients, and training the predictive model to identify patterns in the values corresponding to myocardial infarction occurrence reflected in the past data.

11. The method of any one of claims 7 to 10, wherein the heart rate sensor is a contactless sensor.15 12. The system of any one of claims 7 to 11, wherein the heart rate sensor is configured for use in a motor vehicle to obtain the heart rate signal.

Images