Method and system for establishing a mortality prediction model for hemodialysis patients, and method and program for predicting mortality in hemodialysis patients

Abstract

To provide a method for establishing a mortality prediction model for hemodialysis patients.SOLUTION: A method to be executed by a processor includes: acquiring a training dataset and a verification dataset both corresponding to a plurality of variables; analyzing the variables in the training dataset by using a single-variable Cox proportional hazards model to select a plurality of potentially-predicted variables affecting a probability of hemodialysis patients dying within half a year or one year and to define all the variables as latent variables; using a multivariable Cox proportional hazards model to construct a prediction model for calculating, for the latent variables, the probability of the hemodialysis patients dying within half a year or one year; evaluating accuracy and correction accuracy of the prediction model; verifying the prediction model by the verification dataset; and selecting a plurality of related variables that have the highest accuracy and correction accuracy and have passed verification, to generate and store it as a mortality prediction model for the hemodialysis patients.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method and system for establishing a prediction model, and more particularly to a method and system for establishing a death prediction model for hemodialysis patients that utilizes developments in big data, and a method and program for predicting death in hemodialysis patients (Method and System for Establishing Death in Hemodialysis Patients Prediction Model, and Method and Program for Predicting Death in Hemodialysis Patients). [Background technology]

[0002] According to 2021 statistics from Taiwan's Ministry of Health and Welfare, there are approximately 88,000 patients with end-stage renal disease (ESRD) in Taiwan who are undergoing long-term hemodialysis. The World Health Organization estimates that there are approximately 4 million patients with end-stage renal disease worldwide, of which 2.8 million are undergoing long-term hemodialysis, a staggering number.

[0003] Furthermore, hemodialysis patients are not limited to the elderly; they often suffer from multiple chronic diseases. Long-term hemodialysis treatment can lead to a variety of problems, including malnutrition, sarcopenia, vascular hardening, and unstable blood pressure. According to the Kidney Disease in Taiwan Annual Report (TWRDS), the mortality rate of kidney dialysis patients is higher than that of healthy individuals of the same age, with an average five-year survival rate of only around 53%. In terms of life expectancy, male patients have an average life expectancy of approximately 71 years, while female patients have an average life expectancy of approximately 75 years—approximately 10 years shorter than average. Therefore, the Taiwan Society of Nephrology (TSN) and nephrologists are actively conducting research to reduce mortality and extend the life expectancy of dialysis patients.

[0004] Hemodialysis patients suffer from hardened blood vessels, weakened cardiac function, and weakened immune systems. If a cardiovascular disease or infection develops during the hemodialysis process, it can lead to sudden death before family members have time to respond, making prevention difficult. Therefore, if a patient's risk of death within six months or one year could be predicted, clinicians could examine the patient's physical condition early and provide timely treatment to prevent death. Traditionally, patients undergoing hemodialysis undergo regular monthly blood tests, which include nutritional indicators such as hemoglobin and albumin, and vascular hardening indicators such as calcium ions, phosphorus ions, and parathyroid hormone. Summary of the Invention [Problem to be solved by the invention]

[0005] However, most studies have only analyzed a single factor in the relationship between these blood biochemical indices and patient mortality risk.In addition, the interpretation of these blood test values ​​is entirely dependent on the experience of the individual physicians who visit the hemodialysis center, and is therefore susceptible to the influence of individual physician experience and is not a standardized interpretation method, making it impossible for hemodialysis centers to effectively predict the mortality risk of dialysis patients.

[0006] Therefore, the present inventors believed that the above drawbacks could be improved, and as a result of extensive research, they came up with the proposal of the present invention, which effectively improves the above problems through rational design.

[0007] The present invention was made in consideration of the above-mentioned problems through intensive research by the inventor, and its main object is to provide a method and system for establishing a mortality prediction model for hemodialysis patients, which predicts the risk of a hemodialysis patient dying within six months or one year by integrating multiple risk factors and performing a comprehensive analysis, and provides the model as a reference for hemodialysis center staff to assist medical staff in making evaluations for subsequent treatment or referral, as well as a method for predicting the mortality of hemodialysis patients. [Means for solving the problem]

[0008] To achieve the above object, one aspect of the present invention provides a method for establishing a mortality prediction model for hemodialysis patients, which is executed by a processor. The method includes the steps of: acquiring a training dataset and a validation dataset, where the training dataset and the validation dataset both correspond to a plurality of variables; analyzing the variables corresponding to the training dataset using a single-variable Cox proportional hazards model to select a plurality of potentially predictive variables for the probability of hemodialysis patients dying within six months or one year, and setting all of these as latent variables; constructing a prediction model for the latent variables using a multivariable Cox proportional hazards model, where the prediction model is used to calculate the probability of hemodialysis patients dying within six months or one year; evaluating the accuracy and corrected accuracy of the prediction model; validating the prediction model using the validation dataset; and selecting and saving a plurality of relevant variables that make the prediction model have the highest accuracy and corrected accuracy and that pass validation as a mortality prediction model for hemodialysis patients.

[0009] Another aspect of the present invention is a system for establishing a mortality prediction model for hemodialysis patients, the system comprising a processor and a storage device connected to the processor and used to store instructions, which, when executed, causes the processor to perform the method for establishing a mortality prediction model for hemodialysis patients of the present invention.

[0010] According to yet another aspect of the present invention, there is provided a method for predicting mortality in a hemodialysis patient, the method being executed by a processor and including the steps of: acquiring actual measurement data of the hemodialysis patient, the actual measurement data including a plurality of actual measurement values ​​of the hemodialysis patient and corresponding to relevant variables of the hemodialysis patient mortality prediction model generated by the method for establishing a mortality prediction model for a hemodialysis patient according to the present invention; and inputting the actual measurement data into the hemodialysis patient mortality prediction model to obtain a predicted probability that the hemodialysis patient will die within six months or one year.

[0011] Furthermore, a further aspect of the present invention is a program that, when executing one or more instructions stored in a computer system, causes a processor to perform the following steps: acquiring actual measurement data of a hemodialysis patient, the actual measurement data including multiple actual measurement values ​​of the hemodialysis patient and corresponding to relevant variables of the hemodialysis patient mortality prediction model generated by the method for establishing a hemodialysis patient mortality prediction model of the present invention; and inputting the actual measurement data into the hemodialysis patient mortality prediction model to obtain a predicted probability that the hemodialysis patient will die within six months or one year. [Effects of the Invention]

[0012] The present invention is configured as described above and therefore provides the following effects. (1) Because it integrates multiple risk factors to comprehensively analyze a patient's risk of death, it is more accurate than an analysis that relies on a single factor or the doctor's personal experience. (2) It can be operated by one person, and the entire prediction process usually takes less than 10 minutes, thereby achieving the effect of saving labor and time. (3) By simply providing basic data, routine disease data, and biochemical blood test data of hemodialysis patients, the probability of the patient dying within six months or one year can be calculated, allowing hemodialysis facilities to take appropriate measures early, thereby improving the quality of care for hemodialysis patients.

[0013] At least the following matters will become clear from the description of this specification and the accompanying drawings. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a flowchart illustrating a method for establishing a mortality prediction model for hemodialysis patients according to one embodiment of the present invention. [Figure 2] 1 is a histogram comparing the difference between predicted and actual mortality rates for patient groups with different mortality risks in the training dataset. DETAILED DESCRIPTION OF THE INVENTION

[0015] The following describes in detail the embodiments of the present invention, but the present invention is not limited to these, and various modifications are possible within the scope of the description, and embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0016] 1 is a flowchart showing a method for establishing a mortality prediction model for hemodialysis patients according to one embodiment of the present invention, which comprises the following steps S1 to S6.

[0017] In step S1, the processor acquires a training dataset and a validation dataset. The training dataset and the validation dataset each correspond to multiple variables, and the data in the training dataset and the validation dataset may be acquired from hemodialysis patients at different hemodialysis facilities. The data collected in the training dataset and the validation dataset may be, for example, all specified by the Taiwan Society of Nephrology and conform to the long-term follow-up data of hemodialysis patients specified by the International Society of Nephrology (ISN), and may be collected using biochemistry blood testing equipment in a typical medical institution's laboratory.

[0018] The variables may also include demographic data, primary renal disease data, co-morbidities, hemogram and biochemical profiles, hemodialysis quality indicators, and time and cause of death of the patient. That is, each of the training data sets in the training dataset includes test values ​​corresponding to the above-mentioned variables, and each of the validation data sets in the validation dataset includes test values ​​corresponding to the above-mentioned variables.

[0019] In step S2, the processor analyzes the variables in the training dataset using a single-variable Cox Proportional Hazard Regression Model to select variables that represent the probability of multiple potentially predicted patients dying within six months or one year, and all of these variables are considered latent variables. If the p-value of a variable in the single-variable Cox Proportional Hazard Regression Model is less than 0.15, it may be considered a latent variable.

[0020] In step S3, the processor constructs a predictive model for these latent variables using a multivariate Cox proportional hazards model, and the predictive model is used to calculate the probability that a hemodialysis patient will die within six months or one year.

[0021] In step S4, the processor evaluates the discrimination and calibration accuracy of the prediction model. The accuracy of the prediction model is evaluated by the area under the curve (AUC) of the receiver operating characteristic curve (ROC curve). If the area under the curve is greater than 0.7, it indicates that the prediction model is worth using. If the area under the curve is greater than 0.8, it indicates that the prediction model has excellent discrimination.

[0022] The difference between the predicted mortality rate and the actual mortality rate for patient groups with different mortality risks in the training dataset is further compared to evaluate the correction accuracy of the prediction model, i.e., evaluate whether the prediction model shows stable performance across all patient groups with high, medium, and low mortality risks.

[0023] In step S5, the processor validates the prediction model using the validation dataset and compares whether the prediction model matches the prediction results of the validation prediction model constructed using the validation dataset. Specifically, a validation prediction model is first constructed using the validation dataset, and the area under the receiver operating characteristic curve of the validation prediction model is calculated. Then, mortality prediction for the validation dataset is performed using the prediction model constructed using these latent variables, and the area under the receiver operating characteristic curve of the validation prediction model is calculated and the two areas under the curve are compared. If the areas under the curve of the two are similar, it indicates that the predictive abilities of the two models are equivalent, and the prediction model passes validation.

[0024] In step S6, the processor selects a number of relevant variables that have the highest accuracy and correction accuracy of the prediction model and that have passed validation, and generates and stores the selected relevant variables as a hemodialysis patient mortality prediction model, which is used to calculate the probability that a hemodialysis patient will die within six months or one year.

[0025] For example, we collected anonymized hemodialysis data from patients with end-stage renal disease who underwent hemodialysis at hospital hemodialysis centers between 2010 and 2018. After eliminating ineligible data, such as data with missing values, data under 20 years old or over 90 years old, data from patients who had undergone kidney transplants, or data that did not meet the necessary criteria, we used the data of approximately 17,359 patients as the training dataset and the data of approximately 13,656 patients as the validation dataset.

[0026] Next, the following relevant variables will be selected to establish a mortality prediction model for hemodialysis patients using a multivariable Cox proportional hazards model. The relevant variables, such as the patient's age, past medical history including atrial fibrillation, coronary arterial disease, peripheral arterial disease, dementia, malignancy, and hospitalization due to infection within the past year, and the patient's usual biochemical blood test data, including creatinine, intact-PTH, high-sensitivity C-reactive protein (CRP), alkaline phosphatase, high-density lipoprotein cholesterol, glycated hemoglobin, albumin, aspartate transaminase, white blood cell count, platelet count, and dialysis efficiency (KT / V), are all patient information that can be obtained at a typical hemodialysis center.

[0027] Regarding accuracy, the mortality prediction model for hemodialysis patients constructed using the above-mentioned related variables had an area under the receiver operating characteristic curve of 0.808, indicating that the prediction model has good accuracy.

[0028] Regarding the correction accuracy, Figure 2 shows histograms comparing the difference between predicted and actual mortality rates for patient groups with different mortality risks in the training dataset. The horizontal axes 1 to 10 in Figure 2 represent the mortality risk in deciles (deciles of risk) from low to high. The vertical axis of the black rectangular plot (Predicted (Cox)) represents the 1-year mortality risk (%Risk of 1-year death) predicted by the prediction model. The vertical axis of the gray rectangular plot (Observed (KM)) represents the actual mortality rate for patients in the training dataset. The prediction model demonstrated stable correction accuracy for all patients at low to high risk, with the highest accuracy being particularly accurate for the one-tenth of patients with the highest mortality rate. This group of patients at high risk of death requires early warning using the hemodialysis patient mortality prediction model established by the present invention. The prediction model also demonstrated the highest correction accuracy for this patient group.

[0029] Furthermore, a validation prediction model was constructed using the validation dataset, and the area under the receiver operating characteristic curve of the validation prediction model was 0.792. When mortality prediction was performed on the validation dataset using the hemodialysis patient mortality prediction model constructed using the related variables, the area under the receiver operating characteristic curve was 0.789, which is equivalent to the predictive ability of a prediction model developed using the validation dataset itself, indicating that the hemodialysis patient mortality prediction model passed validation. It can be recommended to other hemodialysis centers, maintaining stable predictive ability.

[0030] The method for establishing a mortality prediction model for hemodialysis patients according to the present invention utilizes big data to develop a prediction model that can effectively predict the probability of hemodialysis patients dying within six months or one year. Furthermore, the prediction model integrates multiple risk factors to perform a comprehensive analysis of a patient's mortality risk, which is more accurate than analyses based on a single factor or the doctor's personal experience. The prediction model has stable prediction ability across hemodialysis groups with different mortality risks and has passed validation, further demonstrating that the prediction model maintains stable prediction ability even when applied to different hemodialysis centers.

[0031] The present invention further provides a system for establishing a mortality prediction model for hemodialysis patients, the system comprising: a processor; and a storage device connected to the processor and adapted to store instructions, the processor, when executing the instructions, performing the method for establishing a mortality prediction model for hemodialysis patients of the present invention. The instructions may be a software program or machine executable code.

[0032] The present invention also provides a method for predicting mortality in a hemodialysis patient, which is executed by a processor, and includes the steps of: acquiring actual measurement data of the hemodialysis patient, the actual measurement data including a plurality of actual measurement values ​​such as basic data, general disease data, and blood biochemistry test values ​​of the hemodialysis patient, and corresponding to related variables of the hemodialysis patient mortality prediction model generated by the method for establishing a hemodialysis patient mortality prediction model of the present invention; and inputting the actual measurement data into the hemodialysis patient mortality prediction model to obtain a predicted probability that the hemodialysis patient will die within six months or one year.

[0033] In a preferred embodiment of the method for predicting mortality in hemodialysis patients of the present invention, the mortality prediction model for hemodialysis patients is converted into a simplified score system, which is used for clinical application by operators.

[0034] For example, after inputting the actual measurement data of a hemodialysis patient, if the hemodialysis patient mortality prediction model calculates that the patient has a higher than 50% chance of dying within one year, the medical team will promptly examine the patient's biochemical data and, if malnutrition is detected, immediately consult a nutritionist or provide nutritional supplementation during hemodialysis. If the patient's vascular hardening is severe or blood pressure control is poor, the patient will be referred to a cardiovascular surgeon as soon as possible for further testing and treatment. These early prediction and early treatment methods can extend the lifespan of hemodialysis patients. Furthermore, if the patient is elderly or suffers from comorbidities such as dementia, malignant tumors, or multiple organ failure and is expected to have a short life expectancy, the prediction results can help the family prepare early, preventing it from being too late in the event of the patient's sudden death.

[0035] The method for predicting the mortality rate of hemodialysis patients according to the present invention allows a medical staff member, such as a nurse, caregiver, or office worker, to input necessary data and then calculate the probability of death of the hemodialysis patient within six months or one year. Using the computing power of a conventional computer, the entire prediction process can usually be completed in less than 10 minutes, thereby saving time and labor. Furthermore, the prediction result can be provided to hemodialysis facilities for reference, allowing them to take appropriate measures early, thereby improving the quality of care for hemodialysis patients.

[0036] The present invention also provides a program and a non-transitory computer-readable storage medium having the program stored thereon, the program or the non-transitory computer-readable storage medium storing one or more computer-executable instructions and causing a processor to execute the one or more instructions, to perform the following steps in a computer system: acquiring actual measurement data of a hemodialysis patient, the actual measurement data including a plurality of actual measurement values ​​of the hemodialysis patient and corresponding to relevant variables of the hemodialysis patient mortality prediction model generated by the method for establishing a hemodialysis patient mortality prediction model of the present invention; and inputting the actual measurement data into the hemodialysis patient mortality prediction model to obtain a predicted probability that the hemodialysis patient will die within six months or one year. That is, when the program or non-transitory computer-readable storage medium executes one or more instructions stored in a computer system, the program or non-transitory computer-readable storage medium causes a processor to perform the steps of: acquiring actual measurement data of a hemodialysis patient, the actual measurement data including a plurality of actual measurement values ​​of the hemodialysis patient and corresponding to relevant variables of the hemodialysis patient mortality prediction model generated by the method for establishing a hemodialysis patient mortality prediction model of the present invention; and inputting the actual measurement data into the hemodialysis patient mortality prediction model to obtain a predicted probability that the hemodialysis patient will die within six months or one year. The non-transitory computer-readable storage medium may be a hard disk, an optical disk, a USB memory, a ROM or flash memory, a database accessible from a network, or a computer-readable recording medium having the same function.

[0037] In summary, the present invention has the following beneficial effects. (1) Because it integrates multiple risk factors to comprehensively analyze a patient's risk of death, it is more accurate than an analysis that relies on a single factor or the doctor's personal experience. (2) It can be operated by one person, and the entire prediction process usually takes less than 10 minutes, achieving the effect of saving labor and time. (3) By simply providing basic data, routine disease data, and recent biochemical blood test data of hemodialysis patients, the probability of the patient dying within six months or one year can be calculated, allowing hemodialysis facilities to take appropriate action early, thereby improving the quality of care for hemodialysis patients.

[0038] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0039] S1 Step S2 Step S3 Step S4 Step S5 Step S6 Step

Claims

1. 1. A method for establishing a mortality prediction model for hemodialysis patients executed by a processor, comprising: obtaining a training data set and a validation data set, the training data set and the validation data set both corresponding to a plurality of variables; A univariate Cox proportional hazards model was used to estimate the risk of HIV infection from the training data set. A step of analyzing the variables from the above to select variables of the probability that a plurality of potentially predicted hemodialysis patients will die within six months or one year, and setting all of them as latent variables; A predictive model was constructed for these latent variables using a multivariate Cox proportional hazards model. wherein the predictive model is used to calculate the probability that a hemodialysis patient will die within six months or one year; evaluating the accuracy and correction accuracy of the prediction model; validating the predictive model with the validation dataset; A method for establishing a mortality prediction model for hemodialysis patients, comprising the steps of: selecting a plurality of relevant variables that have passed validation and that result in the prediction model having the highest accuracy and corrected accuracy; and generating and saving the selected relevant variables as a mortality prediction model for hemodialysis patients.

2. The training data set is matched by the univariate Cox proportional hazards model. The step of analyzing these variables to select variables of the probability that a hemodialysis patient will potentially die within six months or one year and setting all of them as latent variables includes the step of setting variables with a p-value of less than 0.15 as latent variables in the single-variable Cox proportional hazards model. The method for establishing a mortality prediction model for hemodialysis patients according to claim 1, comprising:

3. The method for establishing a mortality prediction model for hemodialysis patients described in claim 1, characterized in that the step of evaluating the accuracy and correction accuracy of the prediction model includes a step of evaluating the accuracy of the prediction model using the area under the receiver operating characteristic curve, comparing the difference between the predicted mortality rate and the actual mortality rate of patient groups with different mortality risks in the training dataset, and evaluating the correction accuracy of the prediction model.

4. Validating the predictive model with the validation dataset includes: constructing a validation prediction model using the validation dataset and calculating the area under the receiver operating characteristic curve of the validation prediction model; A step of predicting mortality in the validation dataset using the prediction model constructed using these latent variables and calculating the area under the receiver operating characteristic curve; 2. The method for establishing a mortality prediction model for hemodialysis patients according to claim 1, further comprising a step of comparing the areas under the curves of both.

5. The method for establishing a mortality prediction model for hemodialysis patients according to claim 1, characterized in that the data in the training dataset and the validation dataset are respectively obtained from hemodialysis patients at different hemodialysis facilities.

6. 2. The method for establishing a mortality prediction model for hemodialysis patients as described in claim 1, characterized in that the variables of the training dataset and the validation dataset include demographic data, primary renal disease data, comorbidities, hemogram and biochemical profiles, blood dialysis quality indicators, and patient time and cause of death.

7. A system for establishing a mortality prediction model for hemodialysis patients, comprising: a processor; a storage device coupled to the processor and adapted to store instructions, the storage device causing the processor to: obtaining a training data set and a validation data set, the training data set and the validation data set both corresponding to a plurality of variables; The training data set was compared using a univariate Cox proportional hazards model. A step of analyzing the variables from the above to select variables of the probability that a plurality of potentially predicted hemodialysis patients will die within six months or one year, and setting them all as latent variables; A predictive model was constructed for these latent variables using a multivariate Cox proportional hazards model. the prediction model is used to calculate the probability that a hemodialysis patient will die within six months or one year; evaluating the accuracy and correction accuracy of the prediction model; validating the predictive model with the validation dataset; A system for establishing a mortality prediction model for hemodialysis patients, characterized by performing the steps of: selecting multiple relevant variables that ensure the prediction model has the highest accuracy and corrected accuracy and that have passed verification, and generating and saving them as a mortality prediction model for hemodialysis patients.

8. the instructions cause the processor, upon executing the instructions, to generate a Cox proportional hazards model using the univariate Cox proportional hazards model. In this case, a step of selecting variables with a p-value of less than 0.15 as potential variables is further performed.

8. A system for establishing a mortality prediction model for hemodialysis patients as described in claim 7.

9. The system for establishing a mortality prediction model for hemodialysis patients described in claim 7, characterized in that when executing the instructions, the processor further performs the steps of evaluating the accuracy of the prediction model using the area under the receiver operating characteristic curve, comparing the difference between the predicted mortality rate and the actual mortality rate for patient groups with different mortality risks in the training dataset, and evaluating the correction accuracy of the prediction model.

10. causing the processor, when executing the instructions, to: constructing a validation prediction model using the validation dataset and calculating the area under the receiver operating characteristic curve of the validation prediction model; A step of predicting mortality in the validation dataset using the prediction model constructed using these latent variables and calculating the area under the receiver operating characteristic curve; 8. The system for establishing a mortality prediction model for hemodialysis patients according to claim 7, further comprising the step of comparing the areas under the curves of both.

11. The system for establishing a mortality prediction model for hemodialysis patients described in claim 7, characterized in that the data in the training dataset and the validation dataset are respectively obtained from hemodialysis patients at different hemodialysis facilities.

12. The system for establishing a mortality prediction model for hemodialysis patients described in claim 7, characterized in that the variables of the training dataset and the validation dataset include demographic data, primary renal disease data, comorbidities, hemogram and biochemical profiles, blood dialysis quality indicators, and the patient's time and cause of death.

13. 1. A method for predicting mortality in a hemodialysis patient executed by a processor, comprising: a step of acquiring actual measurement data of a hemodialysis patient, the actual measurement data including a plurality of actual measurement values ​​of the hemodialysis patient and corresponding to related variables of the mortality prediction model for the hemodialysis patient generated by the method for establishing a mortality prediction model for a hemodialysis patient according to any one of claims 1 to 6; A method for predicting death in a hemodialysis patient, comprising the steps of: inputting the actual measured data into a death prediction model for the hemodialysis patient, and obtaining a predicted probability that the hemodialysis patient will die within six months or one year.

14. a processor in executing one or more instructions stored in the computer system; a step of acquiring actual measurement data of a hemodialysis patient, the actual measurement data including a plurality of actual measurement values ​​of the hemodialysis patient and corresponding to related variables of the mortality prediction model for the hemodialysis patient generated by the method for establishing a mortality prediction model for a hemodialysis patient according to any one of claims 1 to 6; A program characterized by executing a step of inputting the actual measurement data into a mortality prediction model for the hemodialysis patient and obtaining a predicted probability that the hemodialysis patient will die within six months or one year.

Images