A method and system for enabling the prediction of a health risk condition
The method and system enhance cardiac surgery risk prediction by employing a stacking ensemble of AI models with synthetic sample generation and feature selection, addressing data imbalance and improving prediction accuracy for personalized and real-time clinical decision-making.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- ISTINYE UNIVERSITESI
- Filing Date
- 2025-12-04
- Publication Date
- 2026-06-25
AI Technical Summary
Current risk prediction models for cardiac surgery, such as EuroSCORE and STS, overestimate mortality risk in high-risk patients, leading to unreliable predictions and incorrect calibration, and machine learning models face challenges like limited datasets, high computational costs, and deficiencies in calibration and discrimination, limiting their reliability in clinical applications.
A method and system using a stacking ensemble of AI models, trained with synthetic sample generation and feature selection techniques like ADASYN and Gini importance, to improve prediction accuracy by balancing data and selecting critical variables, combined with the ERES model for enhanced decision-making.
Provides highly accurate and reliable mortality and complication risk predictions post-cardiac surgery, supporting personalized and real-time clinical decision-making.
Smart Images
Figure TR2025051578_25062026_PF_FP_ABST
Abstract
Description
[0001] A METHOD AND SYSTEM FOR ENABLING THE PREDICTION OF A HEALTH RISK CONDITION
[0002] TECHNICAL FIELD
[0003] The invention relates to a method for estimating a health risk condition that may occur in a patient after cardiac surgery
[0004] PRIOR ART
[0005] Cardiovascular diseases cause the death of more than four million people annually. This situation imposes a high cost on the global economy. Therefore, in order to achieve sustainable development goals and reduce premature deaths caused by non- communicable diseases, it is necessary to implement effective cost policies and interventions.
[0006] Among the most commonly used risk classification models in cardiac surgery are EuroSCORE (European System for Cardiac Operative Risk Evaluation) and STS (Society of Thoracic Surgeons). These models have been developed to estimate the in- hospital mortality risk after cardiac surgery and provide support to clinical decisionmaking processes. However, traditional risk scoring systems such as EuroSCORE tend to overestimate the actual risk, especially in high-risk patients. This situation causes the model to suffer from a lack of discrimination and calibration and reduces the reliability of predictions in high-risk patient groups. As a result, incorrect calibration may present misleading information to patients and their families regarding existing complication and mortality risks, leading to negative outcomes in shared decisionmaking processes. Therefore, it is of great importance to eliminate the calibration deficiencies of current models and to develop alternative methods that can provide more accurate predictions.
[0007] In order to overcome the limitations of current risk prediction models, machine learningbased approaches have the potential to provide more accurate and reliable risk predictions. However, these models also face challenges such as limited datasets, high computational costs, and deficiencies in calibration and discrimination. Furthermore, the inability to fully capture the relationships among complex clinical variables and the lack of sufficient validation methods limit the reliability of the models in clinical applications.
[0008] As a result, all the above-mentioned problems have made it imperative to make an innovation in the relevant technical field.
[0009] SUMMARY OF THE INVENTION
[0010] The present invention relates to a method and system for eliminating the above- mentioned disadvantages and bringing the new advantages to the relevant technical field.
[0011] An object of the invention is to provide a method and system for estimating the mortality risk that may occur in a patient after cardiac surgery.
[0012] Another object of the invention is to provide a method and system for evaluating the complication risks that may occur after surgery in patients who have undergone cardiac surgery.
[0013] Another object of the invention is to provide a method and system for supporting surgeons in decision-making by estimating the personalized mortality risk of patients who have undergone cardiac surgery.
[0014] In order to achieve all the purposes mentioned above and that will emerge from the detailed description below, the present invention is related to a method for estimating a health risk condition that may occur in a patient after cardiac surgery. Accordingly, it comprises the steps of:
[0015] - receiving at least one patient information of a patient via a user interface provided to allow data entry by at least one person,
[0016] - applying the received patient information from the user interface as input to a first artificial intelligence model that has been trained with multiple patient data and the health risk condition corresponding to the patient data, and that provides a health risk condition as output when a patient information is given as input, and obtaining a first health risk condition related to the patient information as output,
[0017] - communicating the first health risk condition to the user interface via a communication unit. Thus, it is ensured that a personalized mortality risk for a patient is obtained.
[0018] A possible embodiment of the invention is characterized in that after the step of obtaining the first health risk condition as output, it comprises the steps of:
[0019] - applying the patient information received through the user interface as input to a second artificial intelligence model that has been trained with multiple patient data and that provides a health risk condition as output when a patient information is given as input, and obtaining a second health risk condition related to the patient information as output,
[0020] - combining the first health risk condition and the second health risk condition using a stacking ensemble method to obtain a third health risk condition,
[0021] - communicating the third health risk condition to the user interface via the communication unit.
[0022] A possible embodiment of the invention is characterized in that it comprises the step of training the first artificial intelligence model with multiple patient data processed using a synthetic sample generation method.
[0023] A possible embodiment of the invention is characterized in that it comprises the step of training the second artificial intelligence model with multiple patient data processed using a synthetic sample generation method.
[0024] A possible embodiment of the invention is characterized in that it comprises the step of training the first artificial intelligence model with selected patient data chosen from multiple patient data using the Gini importance method in order to obtain the first health risk condition as output.
[0025] The invention also relates to a system for estimating a health risk condition that may occur in a patient after cardiac surgery. Accordingly, its novelty lies in that the above- mentioned method steps are performed by a processor unit. BRIEF DESCRIPTION OF DRAWINGS
[0026] Fig. 1 is a showing representative view operating scenario of the invention.
[0027] DETAILED DESCRIPTION OF THE INVENTION
[0028] In this detailed description, the subject of the invention is explained by way of example only for a better understanding of the subject, which shall not create any limiting effect.
[0029] The invention relates to a method and system for estimating the health risk condition that may occur in a patient after cardiac surgery. In a possible embodiment of the invention, the health risk condition may be the mortality risk, which is well known in the art. In a possible embodiment of the invention, the estimation of the health risk condition after cardiac surgery is provided for adult patients (+18).
[0030] In a possible embodiment of the invention, the cardiac surgery may be a high-risk surgery such as coronary bypass surgery and / or heart valve surgery.
[0031] In order to estimate the health risk condition of the patient after cardiac surgery, first, at least one patient information of the patient is obtained through a user interface (100) provided to allow data entry by at least one person. In a possible embodiment of the invention, said user interface (100) may be a mobile application, a website, etc., accessed via mobile devices such as a computer, tablet, mobile phone, etc.
[0032] In a possible embodiment of the invention, the patient information may include at least one of the patient's demographic information, medical history information, clinical value information, and comorbidity information. The medical history information may include data reflecting the patient's past health status, such as previous illnesses, surgeries, medications used, allergies, and genetic health problems in the family.
[0033] The clinical value information may include biological or physiological data such as creatinine, blood glucose, and blood pressure, which are obtained as a result of tests or measurements performed to evaluate the patient's health status. After the patient information is obtained, the patient information received through the user interface (100) is applied as input to a first artificial intelligence model, which has been trained with multiple patient data and the corresponding health risk conditions, and which provides a health risk condition as output when a patient information is given as input. Thus, the health risk condition of the patients whose information is entered through the user interface (100) is estimated. In a possible embodiment of the invention, the first artificial intelligence model may be the Random Forest (RF) model, which is well known in the art.
[0034] In a possible embodiment of the invention, the first artificial intelligence model may be trained with multiple patient data processed using a synthetic sample generation method known in the art, namely ADASYN (Adaptive Synthetic Sampling). In this way, the performance of the first artificial intelligence model is improved.
[0035] As well known in the art, the ADASYN method is a technique used to increase the number of samples belonging to underrepresented groups in imbalanced datasets. For this purpose, ADASYN generates synthetic samples for underrepresented health risk groups in datasets such as patient information and the corresponding health risk condition. For example, if there are many data points for low health risk but few for high health risk, ADASYN adds synthetic data to the high-risk group to balance the model. In this way, the artificial intelligence is enabled to learn and predict all risk groups more effectively.
[0036] In another possible embodiment of the invention, the first artificial intelligence model may be trained with multiple selected patient data chosen from multiple patient data using the Gini importance method. As well known in the art, the Gini importance method is a technique that enables the measurement of how much each feature contributes to the model’s ability to make accurate predictions. For example, an increase in creatinine levels leads to impaired kidney function and the accumulation of toxins and metabolic waste in the body. This increases the workload of other organs, especially the heart. In addition, kidney failure frequently causes fluid and electrolyte imbalances. Therefore, it may lead to arrhythmias and circulatory problems in the heart. All these factors together increase the risk of postoperative complications and the mortality rate. Thanks to Gini importance, since it has been identified that the creatinine level reflects kidney function and therefore has a strong correlation with mortality after cardiac surgery, the first artificial intelligence model makes decisions regarding the health risk condition by using the creatinine level as one of the most important variables, as the first artificial intelligence model. In this way, by highlighting the critical variables, the model is enabled to make more accurate and reliable health risk predictions.
[0037] In a possible embodiment of the invention, after the features selected by Gini importance, the recursive feature elimination with cross-validation (RFECV) method may be used to determine the optimal number of features that will improve the performance of the artificial intelligence model. The recursive feature elimination with cross-validation (RFECV) method enables the automatic determination of the number of features that provide the best performance by using cross-validation. In this way, by eliminating unnecessary features or features that negatively affect the performance of the artificial intelligence model, the artificial intelligence models are enabled to produce more efficient and more accurate results.
[0038] As a result, in datasets, for example, the low number of high-risk patients may negatively affect the performance of existing risk prediction models. For this reason, this problem is addressed using data balancing techniques (e.g., ADASYN) and feature selection methods (e.g., RFECV, Gini importance), thereby improving the overall performance of the model. In this way, artificial intelligence models are able to make more accurate predictions not only in specific patient groups but also in all patient groups.
[0039] The first health risk condition is communicated to the user interface (100) via a communication unit (200). In a possible embodiment of the invention, the communication unit (200) may include a Bluetooth module, a modem, a Wi-Fi card, a mobile data network, etc. In this way, the health risk condition of the patient whose patient information has been entered can be viewed and examined by authorized persons.
[0040] In a possible embodiment of the invention, after the step of obtaining the first health risk condition as output, the patient information received through the user interface (100) is applied as input to a second artificial intelligence model, which has been trained with multiple patient data and provides a health risk condition as output when a patient information is given as input, and thereby a second health risk condition regarding the patient information is obtained as output. In a possible embodiment of the invention, the second artificial intelligence model may be the XGBoost model, which is well known in the art.
[0041] In a possible embodiment of the invention, the second artificial intelligence model may be trained with multiple patient data processed using a synthetic sample generation method known in the art, namely ADASYN (Adaptive Synthetic Sampling). In this way, the performance of the second artificial intelligence model is improved.
[0042] In a possible embodiment of the invention, the first health risk condition and the second health risk condition are combined using the stacking ensemble method, and a third health risk condition is obtained. The obtained third health risk condition is communicated to the user interface (100) via the communication unit (200).
[0043] In a possible embodiment of the invention, an ERES model is created using an ensemble-based risk estimation system (ERES). As well known in the art, ERES is a system that aims to make more accurate and reliable predictions by combining multiple artificial intelligence models. In this way, the problem of data imbalance is addressed, and the prediction accuracy is improved through the combination of different artificial intelligence models. The ERES model is used for mortality prediction after surgery, enabling highly accurate and real-time decision-making in clinical decision processes.
[0044] All steps performed to estimate the health risk condition of a patient may be carried out by a processor unit (300). In a possible embodiment of the invention, the processor unit (300) may be a CPU, a GPU, a microprocessor, etc.
[0045] In a possible embodiment of the invention, the patient information, the first health risk condition, the second health risk condition, and the third health risk condition are recorded in a memory unit (400). In a possible embodiment of the invention, the memory unit (400) may include a combination of memories that permanently store data and also temporarily store data when necessary. The scope of protection of the invention is specified in the appended claims and cannot be limited to what is described for illustrative purposes in this detailed description. It is clear that a person skilled in the art can produce similar embodiments in the light of what is explained above, without deviating from the main theme of the invention.
[0046] REFERENCE NUMERALS GIVEN IN THE DRAWING
[0047] 100 User Interface 200 Communication Unit
[0048] 300 Processor Unit
[0049] 400 Memory Unit
Claims
CLAIMS1 . A method for estimating a health risk condition that may occur in a patient after cardiac surgery, characterized in that it comprises the steps of:- receiving at least one patient information of a patient via a user interface (100) provided to allow data entry by at least one person,- applying the received patient information from the user interface (100) as input to a first artificial intelligence model that has been trained with multiple patient data and the health risk condition corresponding to the patient data, and that provides a health risk condition as output when a patient information is given as input, and obtaining a first health risk condition related to the patient information as output,- communicating the first health risk condition to the user interface (100) via a communication unit (200).
2. A method according to claim 1 , characterized in that after the step of obtaining the first health risk condition as output, it comprises the steps of:- applying the patient information received through the user interface (100) as input to a second artificial intelligence model that has been trained with multiple patient data and that provides a health risk condition as output when a patient information is given as input, and obtaining a second health risk condition related to the patient information as output, combining the first health risk condition and the second health risk condition using a stacking ensemble method to obtain a third health risk condition, communicating the third health risk condition to the user interface (100) via the communication unit (200).
3. A method according to claim 1 , characterized in that it comprises the step of training the first artificial intelligence model with multiple patient data processed using a synthetic sample generation method.
4. A method according to claim 2, characterized in that it comprises the step of training the second artificial intelligence model with multiple patient data processed using a synthetic sample generation method.
5. A method according to claim 1 , characterized in that it comprises the step of training the first artificial intelligence model with selected patient data chosen from multiple patient data using the Gini importance method in order to obtain the first health risk condition as output.
6. A system for estimating a health risk condition that may occur in a patient after cardiac surgery, characterized in that the above-mentioned method steps are performed by a processor unit (300).