RU2650039C1 - Method for prediction of risk of ventricular tachycardia paroxysms development in patients in first days of myocardial infarction - Google Patents

Abstract

FIELD: medicine.

SUBSTANCE: (a) based on the clinical anamnestic data, results of laboratory studies, echocardiography and Holter monitoring data of the patient, a mathematical model of artificial neural networks is built in the form of multi-layer perceptron, consisting of 6 neurons in the input layer (age, heart rate, systolic blood pressure on admission to the hospital, the maximum value of CPK, final diastolic size of the left ventricle and the size of myocardial asynergia), a hidden layer of eight neurons with a hyperbolic tangent activation function in the following form: f(z)=(e^z-e^-z)/(e^z+e^-z), an output layer, consisting of two neurons:

- presence of risk of paroxysmal ventricular tachycardia development and

- absence risk of paroxysmal ventricular tachycardia development, with normalization of the output layer values by Softmax function. When the value of

is greater than or equal to

, a risk of paroxysmal ventricular tachycardia development is concluded, and when the value of

is less than

, ventricular tachycardia development is not predicted.

EFFECT: method allows to accurately predict the development of the paroxysmal ventricular tachycardia, which may contribute to more effective prevention of sudden cardiac death and other life-threatening complications in patients in the early day of myocardial infarction.

3 tbl, 2 ex

Description

The invention relates to the field of medicine, namely to cardiology. It allows predicting the likelihood of developing potentially life-threatening paroxysms of ventricular tachycardia (VT) using an original mathematical model of artificial neural networks (ANNs) and can be used to stratify the risk of sudden cardiac death (SCD) and poor prognosis in patients with myocardial infarction (MI) . The ANN model is built on the basis of the patient's clinical and medical history and laboratory data, as well as the results of Holter monitoring (HM) and echocardiography (echocardiography).

A known method for predicting the severity of arrhythmic syndrome in MI [Logacheva IV, Ponomarev SB, Barantseva NG Patent No. 2518133 RU, cl. A61B 5/0402], which consists in the fact that the patient who underwent MI undergoes daily monitoring of the electrocardiogram (ECG), assesses the standard deviation of the average values of cardio intervals, the ratio of low and high frequency waves and the variability of the QT interval, conducts echo-KG with the determination of the fraction ejection (EF) of the left ventricle (LV). Based on the results obtained, the development of an arrhythmic syndrome is predicted. The method allows to assess the severity of the arrhythmic syndrome with higher accuracy without stress tests.

The disadvantage of this method is the need for additional determination of heart rate variability and QT interval variability in the analysis of XM data.

A known method for predicting the development of SCD and stable VT in patients after MI using the ANN algorithm [Noninvasive prediction of sudden death and sustained ventricular tachycardia after acute myocardial infarction using a neural network algorithm. Zoni-Berisso M., Molini D., Viani S. Ital Heart J. 2001 Aug; 2 (8): 612-20], which consists in the fact that to predict the development of SCD and malignant ventricular arrhythmias (VA) in patients after MI, they use the algorithm of the mathematical model of ANN, which includes such clinical and functional parameters as acute left ventricular failure during the period of the hospital period, the violation of local LV contractility, decreased LVEF, the presence of blockade of the legs of the bundle of His on an ECG, the presence of late ventricular potentials (PCV) and early ventricular repolarization syndrome.

The disadvantage of this method is the need for additional functional studies, in the frequency of the signal-averaged ECG for assessing PCa.

The prototype of the proposed method can be considered a method for determining the likelihood of developing LA in patients aimed at CAG [Predictors of ventricular arrhythmias of high gradations in patients aimed at coronary angiography. Kuznetsov V.A., Yurkina Yu.A., Todosiychuk V.V., et al. Cardiology. - 2014; 8: 44-48], according to which the mathematical method of logistic regression is used as a mathematical model for calculating FA predictors in patients aimed at CAG. Three clinical and functional parameters — LVEF (EF), LV diameter (LV), and patient age — are used as predictors of VA. The likelihood of developing LA (P) is calculated at a separation point of 0.212 and has the form 4.216 + (- 0.048) × EF + 0.082 × LV + 0.032 × age.

The disadvantages of this method can be attributed to: an insufficient number of clinical and functional parameters taken to calculate the likelihood of developing LA, its low specificity (49.5%), and a rather cumbersome calculation formula, which reduces the possibility of its use in everyday medical practice. In addition, the method does not allow to assess the likelihood of development of VT in patients in the acute period of MI.

The aim of the proposed method is the use of the original mathematical model of ANN in predicting the risk of development of VT in patients on the first day of myocardial infarction taking into account clinical and functional parameters.

It is known that in 20% of cases, MI with ST segment elevation is accompanied by the development of life-threatening VA, which are associated with an increase in hospital mortality [Sustained ventricular arrhythmias and mortality among patients with acute myocardial infarction: results from CUSTO-III trial. Al-Khatib S.M., Stebbins A.L., Califf R.M., Lee KL et al. Am Heart J. 2003 Mar; 145 (3): 515-521]. Therefore, special attention is paid to the questions of predicting FAs in the early period of MI.

The prognostic value of VT paroxysms recorded in the hospital period of MI was evaluated in a number of clinical works. So, according to one of the latest studies [Outcome of contemporary acute coronary syndrome complicated by ventricular tachyarrhythmias. Orvin K, Eisen A, Goldenberg I et al. Europace. 2016 Feb; 18 (2): 219-26], the presence of VT and / or ventricular fibrillation (VF) in the hospital period of the disease was associated with a higher level of hospital, 30-day and 1-year mortality. Long-term risk was associated with the presence at the hospital stage of precisely paroxysms of VT, but not VF.

For clinicians, the urgent task is to find factors that can reliably predict the development of VT in patients in the acute period of MI. According to Piccini J.P. et al. [Sustained ventricular tachycardia and ventricular fibrillation complicating non-ST-segment-elevation acute coronary syndromes. Piccini JP, White JA, Mehta RH et al. Circulation. 2012 Jul 3; 126 (1): 41-9] VT and / or VF in patients with myocardial infarction were associated with the development of early heart failure, low LVEF - less than 30%, and the presence of a multivascular lesion of the coronary channel.

Results from another study [Incidence of ventricular arrhythmia and associated patient outcomes in hospitalized acute coronary syndrome patients in Saudi Arabia: findings from the registry of the Saudi Project for Assessment of Acute Coronary Syndrome (SPACE). Hersi AS ¹ , Alhabib KF, AlFaleh HF et al. Ann Saudi Med. 2012 Jul-Aug; 32 (4): 372-7] indicate that the male gender, the presence of signs of acute heart failure (Killip class> 1), systolic blood pressure (SBP) values of less than 90 mm Hg are independent predictors of the development of paroxysms of VT and / or VF in patients with myocardial infarction. In addition, the presence of VT in patients with myocardial infarction according to ChM is associated with higher levels of peak values of CPK and MV-CPK, more frequent occurrence of Q-wave myocardial infarction [Effect of the presence of preinfarction angina pectoris on arrhythmias in the acute period of myocardial infarction: assessment by Holter monitoring. Lykasova E.A., Pak Yu.A., Todosiychuk V.V., Kuznetsov V.A. Bulletin of Arrhythmology 2008; 50: 27-31], which convincingly indicates the relationship of VT with the severity of ischemic reperfusion damage to the myocardium.

Currently, there are no effective methods for identifying and predicting the development of malignant FAs in the population of patients with MI, therefore, one of the most important tasks facing a modern clinician is stratification of the risk of developing life-threatening FAs. Risk stratification refers to the isolation of a complex of factors associated with an increased likelihood of developing any complications [Elliot P. Investigation and treatment of hypertrophic cardiomyopathy. Clin Med 2007; 7: 383-7]. Existing effective methods of preventing SCD in patients with VT would extend the life of these patients and fundamentally change their prognosis. One of the most reliable and reproducible methods to detect the presence, determine the nature and quantity of VT is XM - a long-term, usually within a day, digital recording of an ECG on a portable recorder with subsequent medical analysis of the results. However, XM on the first day of MI is not always available. In this regard, the urgent task is to identify clinical and functional parameters that could serve as predictors of VT development in patients with MI.

Obviously, to solve such difficult formalized problems, which, in particular, reduce many of the problems of medicine, modern neural network technologies can be used. This is due to the fact that the researcher is often provided with a large amount of heterogeneous factual material for which a mathematical model has not yet been created. One of the most convenient tools for solving such problems is ANN - a powerful and at the same time flexible method for simulating processes and phenomena [Oravin O.I. The use of artificial neural networks for the analysis of pathology of blood vessels. Russian Journal of Biomechanics. - 2011; 3: 45-51]. The basis of each ANN is simple elements called artificial neurons, which mimic the functioning of brain neurons.

The proposed method for predicting the risk of VT development in patients with myocardial infarction was developed during our research using the mathematical model of ANN. The study included 53 patients with myocardial infarction (40 men, mean age 55.9 ± 11.4 years). All patients underwent a comprehensive clinical laboratory and echocardiographic examination according to current recommendations [Schiller N.B., Shah P.M., Crawford M. et al. Recommendation for quantitation of the left ventricle by two-dimentional echocardiography // Arch. Inst. Cardiol. Fur.- 1984.- Vol. 54 (4). - P. 405-409], as well as 24-hour CM in the first day of the disease.

VT was defined as 3 or more ventricular complexes with a heart rate of more than 100 beats. per min [ACC / AHA / ESC guidelines for management of patients with ventricular arrhythmias and the prevention of sudden cardiac death-executive summary. Eur Heart J. 2006; 27: 2099-2140].

According to the results of HM VT, 39 (73.5%) patients were identified. We evaluated: gender, age, heart rate (HR), systolic and diastolic blood pressure (SBP and DBP), history of arterial hypertension and diabetes mellitus, maximum values of creatine phosphokinase (CPK) and its MV fraction, as well as echo KG.

To predict the development of VT, the construction of the mathematical model of the ANN was used [Khaikin S. Neural networks: full course, 2nd ed. - M .: LLC “I.D. Williams, 2006. - 1004 c.], The type of which can be defined as a multilayer perceptron with one hidden layer.

On the basis of clinical and medical history, laboratory data and the results of echocardiography, a coding table for the input ANN input variables was generated (table 1).

The rationing of quantitative indicators of the input layer is carried out according to the following formulas (1-9):

и рассчитывается по следующим формулам (7-14):The hidden layer is calculated based on the normalized values of neurons and consists of 8 neurons with the activation function of hyperbolic tangent

and is calculated by the following formulas (7-14):

The output layer is calculated based on the values of the neurons of the hidden layer and is represented by two classification indicators obtained after normalizing the values using the Softmax function.

, отсутствие ЖТ -

и рассчитываются по формулам (15) и (17).Classification indicators of the output layer determine the presence or absence of VT: the presence of VT -

, absence of VT -

and are calculated by formulas (15) and (17).

Where

Where

больше

, то по данным модели можно говорить о наличии риска развития пароксизмов ЖТ.If

more

, then according to the model, we can talk about the risk of VT paroxysms.

меньше

, то по данным модели можно говорить об отсутствии риска развития пароксизмов ЖТ.If

smaller

, then according to the model it can be said that there is no risk of VT paroxysms.

To determine the diagnostic accuracy of the obtained model, the initial number of patients was divided into 3 samples: training, test and control (table 2).

Based on the results of constructing the ANN, the sensitivity of the model was analyzed based on the sum of squares of forecasting errors when taking the value of the studied indicator as the average in the total sample (Residual sum of square - RSS). The results of the analysis are presented in table 3.

The smaller the error, the smaller the change in the accuracy of predicting the presence of VT when replacing the studied parameter with its average value.

High-level programming languages such as C #, Java, Object Pascal and others can be used as a tool to implement this ANN model in the form of a software product with a user graphical interface, which automate the calculation of a prognostic indicator and provide the possibility of developing a user interface.

The sensitivity analysis of the ANN model, based on the estimation of the sum of the squares of the forecasting errors, showed that the following clinical and functional parameters showed the greatest significance in predicting the development of VT: heart rate and SBP upon admission, end-diastolic size of the left ventricle, size of myocardial asynergy, maximum values CPK levels and age.

The total diagnostic value of the model is 88.6%, sensitivity is 97.4%, specificity is 64.2%, which indicates its high diagnostic accuracy.

Thus, the task of the proposed method is solved by the fact that according to the method for predicting the risk of VT development in patients with myocardial infarction, on the basis of clinical anamnestic and laboratory data, the results of echocardiography, a mathematical model of the ANN is constructed in the form of a multilayer perceptron consisting of 6 neurons of the input layer, hidden a layer of 8 neurons with an activation function of hyperbolic tangent and an output layer consisting of two neurons with normalization of the Softmax function values.

The essential features characterizing the invention and distinguishing the claimed technical solution from the prototype, showed in the claimed combination of new properties that are not explicitly derived from the prior art and are not obvious to a specialist.

An identical set of features was not found in the patent and medical literature of this and related fields of medicine.

The proposed method has high diagnostic value, sensitivity and specificity, which can significantly improve the quality of diagnostics in a cardiology clinic, and can be recommended for use in practical health care, does not require special conditions, is economical in time, easy to operate.

Based on the foregoing, the proposed invention should be considered relevant to the conditions of patentability "novelty", "inventive step", "industrial applicability".

Clinical example 1

больше

, что позволило определить данного пациента в категорию высокого риска развития ЖТ. Полученные данные подтверждаются результатами ХМ, проведенного в первые сутки заболевания: зарегистрировано 16 пароксизмов неустойчивой мономорфной ЖТ с ЧСС от 116 до 174 ударов в 1 минуту.Patient O., 42 years old, was hospitalized in the clinic of the Tyumen Cardiology Center with a diagnosis of "IHD. Acute transmural (with Q wave) anterolateral myocardial infarction. CHF I FC II (NYHA). ” To stratify the risk of VT development, a mathematical model of ANN was built, in which the following clinical and laboratory data were entered: age - 42 years, LV CRD - 50 mm, asynergy size - 30%, maximum CPK level - 1984 mmol / l, SBP values upon admission - 140 mm Hg, the value of heart rate upon admission - 84 beats / min. The output of the model revealed that the result corresponds to -

more

, which allowed to identify this patient in the category of high risk of VT. The data obtained are confirmed by the results of chemotherapy performed on the first day of the disease: 16 paroxysms of unstable monomorphic VT with heart rate from 116 to 174 beats per minute were recorded.

Clinical example 2

Patient P., 54 years old, was hospitalized at the clinic of the Tyumen Cardiology Center with a diagnosis of "IHD. Acute transmural (from 3.Q) antero-lateral myocardial infarction. CHF I FC II (NYHA).

меньше

, что позволило определить данного пациента в категорию низкого риска развития ЖТ. Полученные данные подтверждаются результатами ХМ: в течение суток ЖТ не регистрируются.To stratify the risk of VT development, a mathematical model of ANN was built, in which the following clinical and laboratory data were entered: age - 54 years, LV CRD - 49 mm, asynergy size - 25%, maximum CPK level - 370 mmol / l, SBP values upon admission - 170 mm Hg, the value of heart rate upon admission - 84 beats / min. At the output of the model in this clinical example, it was revealed that the result obtained corresponds to -

smaller

, which allowed to identify this patient in the category of low risk of VT. The data obtained are confirmed by XM results: VTs are not recorded during the day.

Thus, the proposed method allows to accurately predict the risk of VT in patients with myocardial infarction on the first day of the disease, providing high quality diagnostics, reducing the cost of diagnostic measures and time spent on examination.

Claims (24)

A method for predicting the risk of developing paroxysms of ventricular tachycardia (VT) in patients on the first day of myocardial infarction, which consists in the fact that on the basis of clinical and medical history data: heart rate in beats per minute (HR) upon admission, SBP in mmHg (SAD) upon admission, age in years (Age); laboratory data: the value of the maximum level of KFK in mmol / l (KFK); and the results of an echocardiographic study: the end-diastolic size of the left ventricle in mm (VS), the size of asynergy in% (Asinergia), a mathematical model of an artificial neural network is constructed in the form of a multilayer perceptron, consisting of 6 input layer neurons, 8 hidden layer neurons with activation function hyperbolic tangent

and the output layer with the normalization of values by the Softmax function, while the normalization of the input layer is carried out according to the following formulas (1-6):

the hidden layer of neurons is calculated on the basis of normalized values of the input layer of biochemical parameters of blood serum according to the following formulas (7-14):

the output layer is calculated by the formulas (15-18).

Where

Where

and if the output of the neural network is the value

more value

, we can talk about the risk of developing paroxysms of VT, if less, then we can talk about the lack of risk of developing paroxysms of VT.