Improved group algorithm-based pre-frailty risk prediction system for elderly cancer patients

Abstract

The application discloses a kind of pre-frail risk prediction systems of elderly tumor patients based on improved group algorithm, the system includes: data acquisition module, constructs elderly tumor patient dataset;Data pre-processing module, generates the feature matrix of fusion after elderly tumor patient;Characteristic processing module, forms initial feature set;Feature screening module, initial feature set is input into discrete monarch butterfly optimization algorithm, generates key feature subset;Multi-scale Markov chain modeling module, based on the key feature subset constructs multi-scale Markov chain model, forms risk state sequence;Dynamic risk prediction module, whether the pre-frail risk of elderly tumor patient reaches early warning standard is judged, and generates corresponding risk early warning information.The present application establishes hierarchical prediction mechanism, so that model can be from short-term health fluctuation, medium-term change trend to long-term risk evolution Comprehensive analysis.

Description

Technical Field

[0001] This invention relates to the field of medical technology, and in particular to a system for predicting the pre-deterioration risk of frailty in elderly cancer patients based on an improved population algorithm. Background Technology

[0002] With the accelerating aging of the population, health management and disease prediction for elderly cancer patients have become a key focus in the medical field. The pre-frailty state of cancer patients is a crucial factor affecting their quality of life, treatment tolerance, and disease prognosis. However, the current medical system still faces many challenges in identifying and predicting the pre-frailty state of elderly cancer patients. Traditional assessment methods are insufficient to meet the needs of precision medicine, and there is an urgent need for an efficient and accurate risk prediction method.

[0003] Currently, the assessment of pre-frailty in elderly cancer patients mainly relies on the experience and judgment of clinicians and questionnaire-based assessment systems, such as the Frailty Index and the Mini-Frailty Assessment Scale. While these methods can reflect the patient's health status to some extent, they have the following limitations: First, they are highly subjective, with differences in assessment standards among different doctors leading to inconsistencies in the results. Second, the data sources are relatively limited, mainly relying on the patient's self-report and simple physiological indicators, failing to fully integrate pathology reports, electronic medical records, and long-term follow-up data, thus limiting the accuracy of predictions. Third, traditional methods typically only provide static assessments and lack the ability to dynamically model changes in the patient's health status over time, making early identification and intervention difficult.

[0004] In recent years, with the development of artificial intelligence and data mining technologies, some disease prediction models based on machine learning and big data analysis have begun to be applied in the medical field. These models typically employ machine learning algorithms such as regression analysis, decision trees, and support vector machines to model patient data and improve prediction accuracy. However, traditional methods still face the following problems in application: First, feature selection is insufficient. Many studies rely on manual feature engineering, making it difficult to effectively mine key features from high-dimensional medical data, thus limiting model performance. Second, traditional machine learning methods often struggle to handle time-series data, failing to accurately depict the evolution of frailty in elderly cancer patients, affecting the ability to predict future risks. Third, existing methods have high requirements for model interpretability. While some deep learning models have advantages in prediction accuracy, they lack medical interpretability, making it difficult to provide effective clinical decision support for doctors.

[0005] In summary, existing technologies for predicting pre-frailty risk in elderly cancer patients suffer from several drawbacks, including subjective assessment methods, insufficient data integration, inadequate utilization of temporal information in prediction models, limited feature selection optimization capabilities, and poor model interpretability. To address these issues, a dynamic risk prediction method is needed that can effectively integrate multi-source heterogeneous data, optimize the feature selection process, and incorporate temporal modeling capabilities. This would improve the accuracy and operability of pre-frailty risk prediction in elderly cancer patients, thereby providing more precise support for clinical decision-making. Summary of the Invention

[0006] One objective of this invention is to propose a risk prediction system for pre-frailty in elderly cancer patients based on an improved population algorithm. This invention establishes a hierarchical prediction mechanism, enabling the model to conduct a comprehensive analysis from short-term health fluctuations and medium-term trends to long-term risk evolution.

[0007] According to an embodiment of the present invention, a pre-frailty risk prediction system for elderly cancer patients based on an improved population algorithm includes:

[0008] The data acquisition module collects multidimensional data from elderly cancer patients and constructs a dataset of elderly cancer patients.

[0009] The data preprocessing module preprocesses the dataset of elderly cancer patients to generate a fused feature matrix of elderly cancer patients.

[0010] The feature processing module, based on the feature matrix of elderly cancer patients, uses statistical analysis and medical expert experience to perform preliminary screening of various features in the data of elderly cancer patients, forming an initial feature set;

[0011] The feature selection module inputs the initial feature set into the discrete monarch butterfly optimization algorithm, and uses a global search strategy to optimize and select features from the data of elderly cancer patients, generating a subset of key features;

[0012] The multi-scale Markov chain modeling module constructs a multi-scale Markov chain model based on the key feature subset, models the health status of elderly cancer patients at different time scales, defines the patient state and state transition probability matrix, and forms a risk state sequence.

[0013] The dynamic risk prediction module dynamically predicts the pre-frailty risk of elderly cancer patients based on a risk state sequence and obtains real-time risk assessment results. It compares the real-time risk assessment results with a preset risk warning threshold to determine whether the pre-frailty risk of elderly cancer patients meets the warning criteria and generates corresponding risk warning information.

[0014] A method for predicting the pre-frailty risk of elderly cancer patients based on an improved population algorithm is applied to a system for predicting the pre-frailty risk of elderly cancer patients based on an improved population algorithm, and includes the following steps:

[0015] S1. Collect physiological parameter data, pathology report data, electronic medical record data, and physician evaluation data from elderly cancer patients to construct a dataset of elderly cancer patients;

[0016] S2. Preprocess the dataset of elderly cancer patients to generate a fused feature matrix of elderly cancer patients;

[0017] S3. Based on the feature matrix of elderly cancer patients, statistical analysis and medical expert experience are used to initially screen various features in the data of elderly cancer patients to form an initial feature set;

[0018] S4. Input the initial feature set into the discrete monarch butterfly optimization algorithm, and use a global search strategy to optimize and select features in the data of elderly cancer patients to generate a subset of key features;

[0019] S5. Construct a multi-scale Markov chain model based on the aforementioned key feature subset, model the health status of elderly cancer patients at different time scales, define patient status and state transition probability matrix, and form a risk state sequence;

[0020] S6. Dynamically predict the pre-frailty risk of elderly cancer patients based on the risk status sequence, obtain real-time risk assessment results, compare the real-time risk assessment results with the preset risk warning threshold, determine whether the pre-frailty risk of elderly cancer patients reaches the warning standard, and generate corresponding risk warning information.

[0021] Optionally, S1 includes the following steps:

[0022] S11. Collect physiological parameter data from elderly cancer patients. This includes heart rate, blood pressure, blood oxygen saturation, body temperature, and blood glucose levels;

[0023] S12. Collect pathology report data from elderly cancer patients. This includes tumor histological type, pathological stage, and expression levels of related biomarkers;

[0024] S13. Collect electronic medical record data of elderly cancer patients This includes the patient's basic information, past medical history, family medical history, treatment plan, and medication details;

[0025] S14. Collect physician assessment data for elderly cancer patients This includes subjective assessment scores, frailty scores, functional status scores, and cognitive ability assessments;

[0026] S15. Based on the timestamp information, the collected physiological parameter data, pathology report data, electronic medical record data, and physician evaluation data of elderly cancer patients are time-aligned to construct a dataset of elderly cancer patients:

[0027]

[0028] in, This is the final dataset of elderly cancer patients.

[0029] Optionally, S2 includes the following steps:

[0030] S21. Dataset of elderly cancer patients Perform data format conversion to transform data from different sources into a unified storage format and construct a standardized dataset of elderly cancer patients;

[0031] S22. Perform data cleaning on the standardized elderly cancer patient dataset to remove redundant and duplicate data, forming a cleaned elderly cancer patient dataset.

[0032] S23. Detect outliers in the cleaned elderly cancer patient dataset, and use statistical methods to process the outliers to generate an outlier-processed elderly cancer patient dataset.

[0033] S24. After outlier processing, detect missing data in the elderly cancer patient dataset, and use mean imputation, interpolation imputation, or imputation based on historical data to process the missing data, and obtain the elderly cancer patient dataset after missing data imputation.

[0034] S25. After imputing the missing data, the dataset of elderly cancer patients is normalized so that each feature value is mapped to the same numerical range, thus obtaining the normalized dataset of elderly cancer patients.

[0035] S26. Standardize the normalized dataset of elderly cancer patients so that the data features conform to a standard normal distribution, thus obtaining the standardized dataset of elderly cancer patients. And generate a fused feature matrix of elderly cancer patients. .

[0036] Optionally, S3 includes the following steps:

[0037] S31. From the fused feature matrix of elderly cancer patients Extract all feature variables and construct a feature set for elderly cancer patients:

[0038]

[0039] in, This is a set of characteristics of elderly cancer patients. Indicates the first Feature variables, The total number of features;

[0040] S32. Calculate the feature set of elderly cancer patients The mean of each feature variable and standard deviation Features with variability below a threshold in all patient samples were removed to form a feature set of elderly cancer patients after variability screening. ;

[0041] S33. Calculate the feature set of elderly cancer patients after variability screening. The correlation between various characteristic variables and the frailty status of elderly cancer patients was analyzed. Characteristic variables with correlations below a set threshold were removed. Based on the experience of medical experts, the characteristic set of elderly cancer patients after correlation screening was further refined. A clinical significance assessment was conducted, and medically insignificant characteristic variables were removed to form a characteristic set of elderly cancer patients after medical screening.

[0042]

[0043] in, A set of characteristics of elderly cancer patients after medical screening;

[0044] S34. Calculate the characteristic set of elderly cancer patients after medical screening based on statistical analysis methods. The importance weights of each feature variable are determined, and the top-weighted variables are selected. The characteristic variables ultimately form the initial feature set:

[0045]

[0046] in, This is the initial set of features selected for the final screening.

[0047] Optionally, S4 includes the following steps:

[0048] S41. Based on the initial feature set Constructing an initial discrete monarch butterfly population matrix for predicting pre-frailty risk in elderly cancer patients. :

[0049]

[0050] in, For the first The initial feature selection vector of an individual monarch butterfly. This indicates the first stage of risk prediction for pre-frailty in elderly cancer patients. This indicates whether a feature is selected; 1 means selected, and 0 means not selected. , For population size, The total number of features in the initial feature set;

[0051] The initialization of the discrete monarch butterfly population was based on the statistical distribution of clinical data, which enabled it to cover the characteristic patterns of pre-debilitative elderly cancer patients of different types in its initial state.

[0052] S42. Based on the accuracy, feature importance, and data redundancy of predicting pre-frailty risk in elderly cancer patients, a fitness function is constructed:

[0053]

[0054] in, Indicates based on the first The accuracy of the pre-frailty risk prediction model for elderly cancer patients constructed from a subset of features. This represents the dimension of the selected feature subset. Representing characteristic variables and The correlation between them These are the weighting coefficients;

[0055] Calculate the initial fitness value of the fitness function. It is used to sort the population individuals and determine the initial global optimum. ;

[0056] S43. Calculate the fitness value of all individuals in the discrete monarch butterfly population, and determine the current global optimal solution based on the fitness value. Then, the feature selection vectors of the migrating subgroups are updated, and a hybrid discrete migration strategy is adopted to bring them closer to the optimal solution:

[0057]

[0058] in, This represents the selection state of the i-th feature of the j-th migrating subgroup individual after iteration t+1. This represents the selection state of the globally optimal feature subset on the i-th feature at iteration t. and They are two independent random numbers, For adaptive migration probability, and fitness value Positive correlation This represents the selection state of the i-th feature for the j-th individual in the current migratory subgroup during the t-th iteration;

[0059] S44. For non-migratory subgroups, feature subset optimization is performed using a local perturbation operator guided by medical prior knowledge:

[0060] ;

[0061] in, For local perturbation probability, and Related, For medical experts to identify characteristics Clinical importance assessment value.

[0062] S45. Merge the fitness values ​​of migrating and non-migrating subgroups:

[0063]

[0064] in, The fusion weight depends on the mean fitness of the migrating and non-migrating subgroups. Select vectors for the individual features of monarch butterflies updated after the (t+1)th iteration;

[0065] Calculate the new fitness:

[0066]

[0067] And update the globally optimal individual:

[0068]

[0069] in, Let represent the set of fitness values ​​of all monarch butterfly individuals after the (t+1)th iteration. This represents the monarch butterfly individual with the highest fitness after the (t+1)th iteration, which is the globally optimal feature subset of the current iteration.

[0070] S46. The Discrete Monarch Butterfly optimization algorithm stops when any of the following conditions are met:

[0071] Reaching the maximum number of iterations ;

[0072] Continuous T rounds of optimization, That is, the globally optimal individual no longer changes, and eventually converges to obtain the globally optimal individual:

[0073]

[0074] S47. The final globally optimal individual Corresponding key feature subset :

[0075] .

[0076] Optionally, S5 includes the following steps:

[0077] S51. Based on the aforementioned key feature subset Extract observational data of elderly cancer patients at different time points to construct a time series of their health status:

[0078]

[0079] in, For the key feature matrix at each time point Observations at that location ,in Indicates the first At the first time point Observations of key features;

[0080] S52. Discretize the pre-frailty health status of elderly cancer patients, mapping the patient's status to a finite set of health levels:

[0081]

[0082] in. A set of health states, representing different health states, including normal. Mild weakness Moderate weakness and severe weakness ;

[0083] A threshold segmentation method is used based on the statistical distribution of key feature subsets to segment the observations at each time point. Mapped to a set of health states :

[0084]

[0085] in For the first Health status at a specific point in time Representing observation data Belonging to state The probability of;

[0086] S53. Establish multi-scale Markov chain models at different time scales. :

[0087]

[0088] in, Representing the time scale, establishing short-term... Mid-term and long-term Markov chain model, The state transition probability matrix represents the changing pattern of health status at different time scales.

[0089] S54. Multiscale Markov Chain Model In the short timescale Next, calculate the transition probability of health status:

[0090]

[0091] in, The probability of a short-term health status transition indicates the patient's transition from a certain state to a certain condition. Change to state The probability, and These represent the health status at adjacent time points, and the short-term transition probability is estimated based on follow-up data.

[0092] S55. Multiscale Markov Chain Model In the medium timescale The following calculation is performed using the state accumulation transition method:

[0093]

[0094] in, The probability of transitioning to a new health status in the intermediate stage represents the patient's transition from a certain state. Change to state The probability, It reflects the probability of two-step state transitions and estimates the long-term trend through intermediate transitions between states;

[0095] S56. Multiscale Markov Chain Model On a long time scale Below, the long-term trend is calculated using steady-state probability:

[0096]

[0097] in, The probability of transitioning to a long-term healthy state represents the patient's transition from a certain state. Change to state The probability, It reflects the evolution trend of long-term health status and predicts the long-term health risk of elderly cancer patients by calculating the steady-state distribution of the transfer matrix.

[0098] S57. Based on the state transition probability matrices at different time scales calculated in S54-S56, calculate the patient's risk state sequences in the short, medium, and long term:

[0099]

[0100] in, Indicated on the time scale Below, the patient is Predicting health status in real time To allow patients to transition from their current healthy state over a timescale d. Changes to future health status The state transition probability.

[0101] Optionally, S6 includes the following steps:

[0102] S61. Based on patient risk status sequence To construct an individualized pre-frailty risk probability for elderly cancer patients:

[0103]

[0104] in, Indicated on the time scale Under these circumstances, elderly cancer patients Constantly in the early stages of weakness or more severe conditions, i.e., mild weakness. Moderate weakness and severe weakness The probability of;

[0105] S62. Based on the multi-scale pre-weakness risk probability, and integrating short-term, medium-term, and long-term risks, calculate the global risk trend:

[0106]

[0107] in, Indicates in At any given moment, the probability of early-stage weakness is determined by combining short-term, medium-term, and long-term forecasts. Weights for different time scales;

[0108] S63. Based on the overall risk trend, assess the risk change trend in the early stages of decline:

[0109]

[0110] in, Indicates in At any given time, the patient's pre-deterior risk rate changes; if This indicates an increased risk of deterioration in the patient. This indicates a reduced risk for the patient;

[0111] S64. Pre-set a set of risk warning thresholds:

[0112]

[0113] in, This indicates a low-risk threshold, corresponding to a mild risk of weakening. This represents the medium risk threshold, corresponding to a moderate risk of weakening. This indicates a high-risk threshold, corresponding to a severe weakening risk;

[0114] S65. Assess the global risk probability based on the risk warning threshold. Risk level assessment:

[0115] ;

[0116] in, Indicates in Risk level at any given moment;

[0117] S66. Based on the rate of change of risk Risk level determination Generate personalized dynamic risk warnings:

[0118] ;

[0119] in, Indicates in The system generates dynamic risk warning information in real time. If the patient's risk level is below the threshold and there is no obvious upward trend, the system will not trigger a warning. If the risk level is moderate or the risk is rising, a warning reminder will be triggered. If the risk level is high or the rate of increase exceeds the high-risk threshold, a high-risk alarm will be triggered.

[0120] The beneficial effects of this invention are:

[0121] (1) This invention uses an improved discrete monarch butterfly optimization algorithm for feature selection. By combining a global search strategy with the experience of medical experts, it effectively optimizes the key feature subset required for predicting the risk of pre-deterioration in elderly cancer patients and introduces a hybrid discrete migration strategy. In the feature subset optimization process, individual migration and local perturbation operators are combined to make feature selection take into account both global search capability and local optimization accuracy, avoid feature redundancy, and reduce computational overhead.

[0122] (2) In the prediction process, this invention innovatively combines a multi-scale Markov chain model to model the health status of elderly cancer patients at different time scales, which makes up for the shortcomings of traditional static prediction models that are difficult to capture the trend of changes in the health status of patients. Furthermore, by using short-term, medium-term, and long-term health status transition probability matrices, a hierarchical prediction mechanism is established, enabling the model to conduct a comprehensive analysis from short-term health fluctuations and medium-term change trends to long-term risk evolution.

[0123] (3) In the risk assessment stage, the present invention adopts a multi-scale dynamic risk prediction method. Based on the risk status sequence of elderly cancer patients, it integrates short-term, medium-term and long-term prediction information to construct a global risk trend analysis model. Combined with the preset risk warning threshold, the warning level is dynamically adjusted so that it can predict the risk of future frailty in the early stage when the patient's health status changes, and provide the possibility of early intervention. In addition, the present invention uses dynamic risk change rate calculation and combines risk warning level to accurately trigger personalized risk warning, avoiding false alarms or omissions caused by oversensitivity or lag in traditional methods. Attached Figure Description

[0124] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0125] Figure 1 This is a flowchart of a pre-frailty risk prediction system for elderly cancer patients based on an improved population algorithm, as proposed in this invention. Detailed Implementation

[0126] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0127] refer to Figure 1 A pre-frailty risk prediction system for elderly cancer patients based on an improved population algorithm includes:

[0128] The data acquisition module collects multidimensional data from elderly cancer patients and constructs a dataset of elderly cancer patients.

[0129] The data preprocessing module preprocesses the dataset of elderly cancer patients to generate a fused feature matrix of elderly cancer patients.

[0130] The feature processing module, based on the feature matrix of elderly cancer patients, uses statistical analysis and medical expert experience to perform preliminary screening of various features in the data of elderly cancer patients, forming an initial feature set;

[0131] The feature selection module inputs the initial feature set into the discrete monarch butterfly optimization algorithm, and uses a global search strategy to optimize and select features from the data of elderly cancer patients, generating a subset of key features;

[0132] The multi-scale Markov chain modeling module constructs a multi-scale Markov chain model based on a subset of key features, models the health status of elderly cancer patients at different time scales, defines the patient state and state transition probability matrix, and forms a risk state sequence.

[0133] The dynamic risk prediction module dynamically predicts the pre-frailty risk of elderly cancer patients based on the risk state sequence and obtains real-time risk assessment results. It compares the real-time risk assessment results with the preset risk warning threshold to determine whether the pre-frailty risk of elderly cancer patients meets the warning criteria and generates corresponding risk warning information.

[0134] A method for predicting the pre-frailty risk of elderly cancer patients based on an improved population algorithm is applied to a system for predicting the pre-frailty risk of elderly cancer patients based on an improved population algorithm, and includes the following steps:

[0135] S1. Collect physiological parameter data, pathology report data, electronic medical record data, and physician evaluation data from elderly cancer patients to construct a dataset of elderly cancer patients;

[0136] S2. Preprocess the dataset of elderly cancer patients to generate a fused feature matrix of elderly cancer patients;

[0137] S3. Based on the feature matrix of elderly cancer patients, statistical analysis and medical expert experience are used to initially screen various features in the data of elderly cancer patients to form an initial feature set;

[0138] S4. Input the initial feature set into the discrete monarch butterfly optimization algorithm, and use a global search strategy to optimize and select features in the data of elderly cancer patients to generate a subset of key features;

[0139] S5. Construct a multi-scale Markov chain model based on key feature subsets to model the health status of elderly cancer patients at different time scales, define patient status and state transition probability matrix, and form a risk state sequence;

[0140] S6. Dynamically predict the pre-frailty risk of elderly cancer patients based on the risk status sequence and obtain real-time risk assessment results. Compare the real-time risk assessment results with the preset risk warning threshold to determine whether the pre-frailty risk of elderly cancer patients meets the warning standard and generate corresponding risk warning information.

[0141] In this embodiment, S1 includes the following steps:

[0142] S11. Collect physiological parameter data from elderly cancer patients. This includes heart rate, blood pressure, blood oxygen saturation, body temperature, and blood glucose levels;

[0143] S12. Collect pathology report data from elderly cancer patients. This includes tumor histological type, pathological stage, and expression levels of related biomarkers;

[0144] S13. Collect electronic medical record data of elderly cancer patients This includes the patient's basic information, past medical history, family medical history, treatment plan, and medication details;

[0145] S14. Collect physician assessment data for elderly cancer patients This includes subjective assessment scores, frailty scores, functional status scores, and cognitive ability assessments;

[0146] S15. Based on the timestamp information, the collected physiological parameter data, pathology report data, electronic medical record data, and physician evaluation data of elderly cancer patients are time-aligned to construct a dataset of elderly cancer patients:

[0147]

[0148] in, This is the final dataset of elderly cancer patients.

[0149] In this embodiment, S2 includes the following steps:

[0150] S21. Dataset of elderly cancer patients Perform data format conversion to transform data from different sources into a unified storage format and construct a standardized dataset of elderly cancer patients;

[0151] S22. Perform data cleaning on the standardized elderly cancer patient dataset to remove redundant and duplicate data, forming a cleaned elderly cancer patient dataset.

[0152] S23. Detect outliers in the cleaned elderly cancer patient dataset, and use statistical methods to process the outliers to generate an outlier-processed elderly cancer patient dataset.

[0153] S24. After outlier processing, detect missing data in the elderly cancer patient dataset, and use mean imputation, interpolation imputation, or imputation based on historical data to process the missing data, and obtain the elderly cancer patient dataset after missing data imputation.

[0154] S25. After imputing the missing data, the dataset of elderly cancer patients is normalized so that each feature value is mapped to the same numerical range, thus obtaining the normalized dataset of elderly cancer patients.

[0155] S26. Standardize the normalized dataset of elderly cancer patients so that the data features conform to a standard normal distribution, thus obtaining the standardized dataset of elderly cancer patients. And generate a fused feature matrix of elderly cancer patients. .

[0156] In this embodiment, S3 includes the following steps:

[0157] S31. From the fused feature matrix of elderly cancer patients Extract all feature variables and construct a feature set for elderly cancer patients:

[0158]

[0159] in, This is a set of characteristics of elderly cancer patients. Indicates the first Feature variables, The total number of features;

[0160] S32. Calculate the feature set of elderly cancer patients The mean of each feature variable and standard deviation Features with variability below a threshold in all patient samples were removed to form a feature set of elderly cancer patients after variability screening. ;

[0161] S33. Calculate the feature set of elderly cancer patients after variability screening. The correlation between various characteristic variables and the frailty status of elderly cancer patients was analyzed. Characteristic variables with correlations below a set threshold were removed. Based on the experience of medical experts, the characteristic set of elderly cancer patients after correlation screening was further refined. A clinical significance assessment was conducted, and medically insignificant characteristic variables were removed to form a characteristic set of elderly cancer patients after medical screening.

[0162]

[0163] in, A set of characteristics of elderly cancer patients after medical screening;

[0164] S34. Calculate the characteristic set of elderly cancer patients after medical screening based on statistical analysis methods. The importance weights of each feature variable are determined, and the top-weighted variables are selected. The characteristic variables ultimately form the initial feature set:

[0165]

[0166] in, This is the initial set of features selected for the final screening.

[0167] In this embodiment, S4 includes the following steps:

[0168] S41. Based on the initial feature set Constructing an initial discrete monarch butterfly population matrix for predicting pre-frailty risk in elderly cancer patients. :

[0169]

[0170] in, For the first The initial feature selection vector of an individual monarch butterfly. This indicates the first stage of risk prediction for pre-frailty in elderly cancer patients. This indicates whether a feature is selected; 1 means selected, and 0 means not selected. , For population size, The total number of features in the initial feature set;

[0171] The initialization of the discrete monarch butterfly population was based on the statistical distribution of clinical data, which enabled it to cover the characteristic patterns of pre-debilitative elderly cancer patients of different types in its initial state.

[0172] S42. Based on the accuracy, feature importance, and data redundancy of predicting pre-frailty risk in elderly cancer patients, a fitness function is constructed:

[0173]

[0174] in, Indicates based on the first The accuracy of the pre-frailty risk prediction model for elderly cancer patients constructed from a subset of features. This represents the dimension of the selected feature subset. Representing characteristic variables and The correlation between them These are the weighting coefficients;

[0175] Calculate the initial fitness value of the fitness function. It is used to sort the population individuals and determine the initial global optimum. ;

[0176] S43. Calculate the fitness value of all individuals in the discrete monarch butterfly population, and determine the current global optimal solution based on the fitness value. Then, the feature selection vectors of the migrating subgroups are updated, and a hybrid discrete migration strategy is adopted to bring them closer to the optimal solution:

[0177]

[0178] in, This represents the selection state of the i-th feature of the j-th migrating subgroup individual after iteration t+1. This represents the selection state of the globally optimal feature subset on the i-th feature at iteration t. and They are two independent random numbers, For adaptive migration probability, and fitness value Positive correlation This represents the selection state of the i-th feature for the j-th individual in the current migratory subgroup during the t-th iteration;

[0179] S44. For non-migratory subgroups, feature subset optimization is performed using a local perturbation operator guided by medical prior knowledge:

[0180] ;

[0181] in, For local perturbation probability, and Related, For medical experts to identify characteristics Clinical importance assessment value.

[0182] S45. Merge the fitness values ​​of migrating and non-migrating subgroups:

[0183]

[0184] in, The fusion weight depends on the mean fitness of the migrating and non-migrating subgroups. Select vectors for the individual features of monarch butterflies updated after the (t+1)th iteration;

[0185] Calculate the new fitness:

[0186]

[0187] And update the globally optimal individual:

[0188]

[0189] in, Let represent the set of fitness values ​​of all monarch butterfly individuals after the (t+1)th iteration. This represents the monarch butterfly individual with the highest fitness after the (t+1)th iteration, which is the globally optimal feature subset of the current iteration.

[0190] S46. The Discrete Monarch Butterfly optimization algorithm stops when any of the following conditions are met:

[0191] Reaching the maximum number of iterations ;

[0192] Continuous T rounds of optimization, That is, the globally optimal individual no longer changes, and eventually converges to obtain the globally optimal individual:

[0193]

[0194] S47. The final globally optimal individual Corresponding key feature subset :

[0195] .

[0196] In this embodiment, S5 includes the following steps:

[0197] S51. Based on key feature subsets Extract observational data of elderly cancer patients at different time points to construct a time series of their health status:

[0198]

[0199] in, For the key feature matrix at each time point Observations at that location ,in Indicates the first At the first time point Observations of key features;

[0200] S52. Discretize the pre-frailty health status of elderly cancer patients, mapping the patient's status to a finite set of health levels:

[0201]

[0202] in. A set of health states, representing different health states, including normal. Mild weakness Moderate weakness and severe weakness ;

[0203] A threshold segmentation method is used based on the statistical distribution of key feature subsets to segment the observations at each time point. Mapped to a set of health states :

[0204]

[0205] in For the first Health status at a specific point in time Representing observation data Belonging to state The probability of;

[0206] S53. Establish multi-scale Markov chain models at different time scales. :

[0207]

[0208] in, Representing the time scale, establishing short-term... Mid-term and long-term Markov chain model, The state transition probability matrix represents the changing pattern of health status at different time scales.

[0209] S54. Multiscale Markov Chain Model In the short timescale Next, calculate the transition probability of health status:

[0210]

[0211] in, The probability of a short-term health status transition indicates the patient's transition from a certain state to a certain condition. Change to state The probability, and These represent the health status at adjacent time points, and the short-term transition probability is estimated based on follow-up data.

[0212] S55. Multiscale Markov Chain Model In the medium timescale The following calculation is performed using the state accumulation transition method:

[0213]

[0214] in, The probability of transitioning to a new health status in the intermediate stage represents the patient's transition from a certain state. Change to state The probability, It reflects the probability of two-step state transitions and estimates the long-term trend through intermediate transitions between states;

[0215] S56. Multiscale Markov Chain Model On a long time scale Below, the long-term trend is calculated using steady-state probability:

[0216]

[0217] in, The probability of transitioning to a long-term healthy state represents the patient's transition from a certain state. Change to state The probability, It reflects the evolution trend of long-term health status and predicts the long-term health risk of elderly cancer patients by calculating the steady-state distribution of the transfer matrix.

[0218] S57. Based on the state transition probability matrices at different time scales calculated in S54-S56, calculate the patient's risk state sequences in the short, medium, and long term:

[0219]

[0220] in, Indicated on the time scale Below, the patient is Predicting health status in real time To allow patients to transition from their current healthy state over a timescale d. Changes to future health status The state transition probability.

[0221] In this embodiment, S6 includes the following steps:

[0222] S61. Based on patient risk status sequence To construct an individualized pre-frailty risk probability for elderly cancer patients:

[0223]

[0224] in, Indicated on the time scale Under these circumstances, elderly cancer patients Constantly in the early stages of weakness or more severe conditions, i.e., mild weakness. Moderate weakness and severe weakness The probability of;

[0225] S62. Based on the multi-scale pre-weakness risk probability, and integrating short-term, medium-term, and long-term risks, calculate the global risk trend:

[0226]

[0227] in, Indicates in At any given moment, the probability of early-stage weakness is determined by combining short-term, medium-term, and long-term forecasts. Weights for different time scales;

[0228] S63. Based on the overall risk trend, assess the risk change trend in the early stages of decline:

[0229]

[0230] in, Indicates in At any given time, the patient's pre-deterior risk rate changes; if This indicates an increased risk of deterioration in the patient. This indicates a reduced risk for the patient;

[0231] S64. Pre-set a set of risk warning thresholds:

[0232]

[0233] in, This indicates a low-risk threshold, corresponding to a mild risk of weakening. This represents the medium risk threshold, corresponding to a moderate risk of weakening. This indicates a high-risk threshold, corresponding to a severe weakening risk;

[0234] S65. Assess the global risk probability based on the risk warning threshold. Risk level assessment:

[0235] ;

[0236] in, Indicates in Risk level at any given moment;

[0237] S66. Based on the rate of change of risk Risk level determination Generate personalized dynamic risk warnings:

[0238] ;

[0239] in, Indicates in The system generates dynamic risk warning information in real time. If the patient's risk level is below the threshold and there is no obvious upward trend, the system will not trigger a warning. If the risk level is moderate or the risk is rising, a warning reminder will be triggered. If the risk level is high or the rate of increase exceeds the high-risk threshold, a high-risk alarm will be triggered.

[0240] Example 1:

[0241] On June 15, 2023, in the oncology ward of a tertiary hospital in a certain city, Mr. Li, 70 years old, underwent hepatic artery chemoembolization treatment for liver cancer. He also had a history of chronic diseases such as hypertension and diabetes. Upon admission, Mr. Li's physical condition was relatively good, and he was able to walk independently. However, the doctor noticed that his weight had dropped by 5.2 kilograms in the past three months, and he had recently felt weak. According to the hospital's previous assessment procedures, the doctor used the Fried Frailty Index (FI) to score his frailty status. The result showed that Mr. Li's FI score was 3 (mild frailty), but the doctor could not determine whether his health condition would deteriorate further in the coming months.

[0242] Because the hospital recently introduced a pre-frailty risk prediction system for elderly cancer patients based on an improved population algorithm, the doctors decided to use the system to conduct a more accurate frailty risk assessment for Mr. Li.

[0243] The system first collects Mr. Li's multidimensional health data, including:

[0244] Physiological parameters (June 15): Heart rate 82 beats / min, blood pressure 138 / 85 mmHg, blood oxygen saturation 96%, body temperature 36.7℃, fasting blood glucose 7.2 mmol / L;

[0245] Pathological data (June 10): AFP (alpha-fetoprotein) 320 ng / mL, tumor pathology report indicates stage II liver cancer, imaging shows enlarged local lesions in the liver;

[0246] Electronic medical record data: 8 years of history of hypertension, 12 years of history of diabetes, and has been taking aspirin, metformin, and antihypertensive drugs for the past 6 months;

[0247] Doctor's assessment (June 15): Grip strength 21kg, 6-meter walking time 6.8s, slightly unsteady gait, and decreased ability to move independently.

[0248] These data enter the data preprocessing module, and after missing value imputation, data standardization, and outlier detection, the system generates Mr. Li's standardized feature matrix.

[0249] The system then used an improved discrete monarch butterfly optimization algorithm to filter the data for features. During the calculation process, the algorithm prioritized highly correlated features and eliminated redundant variables, ultimately selecting 10 key features, including serum albumin level, blood glucose fluctuation range, walking speed, grip strength, hemoglobin, BMI, C-reactive protein, muscle mass index, heart rate variability, and blood pressure fluctuation.

[0250] In traditional methods, doctors often focus only on grip strength, walking speed, and BMI, while ignoring inflammatory factors (C-reactive protein) and blood sugar fluctuations, which are crucial indicators of frailty.

[0251] Based on the selected features, the system uses a multi-scale Markov chain model (MSMC) to predict the state of frailty and assesses Mr. Li’s health status on short-term (1 month), medium-term (3 months) and long-term (6 months) time scales.

[0252] Short-term forecast (July 15): The system predicts that Mr. Li is still in a state of "mild weakness" and his health status has not changed significantly.

[0253] Mid-term forecast (September 15): The system predicts that Mr. Li's health condition will deteriorate to "moderate frailty", with the frailty index increasing by 15%, walking speed decreasing to 8.1s / 6m, and grip strength decreasing to 19kg.

[0254] Long-term prediction (December 15): The system predicts that Mr. Li's health status will deteriorate to "severe weakness", walking speed will decrease to 9.5s / 6m, grip strength will decrease to 17kg, serum albumin will decrease to 30g / L, diabetes control will worsen, blood sugar fluctuation range will increase, and BMI will further decrease.

[0255] In contrast, the traditional FI scoring method assessed Mr. Li as "mildly frail" on June 15, but could not predict his health changes in the coming months.

[0256] Based on the prediction results, the system triggered a "medium risk" alert on July 15, reminding doctors to pay attention to Mr. Li's health status and recommending increased nutritional support, muscle strength training, and gait improvement training interventions.

[0257] On September 15, the system automatically analyzed Mr. Li's new data and found that his BMI had further decreased to 21.5, his C-reactive protein had increased to 8.5 mg / L, and his walking speed had decreased to 8.2 s / 6 m. The system then triggered a "high-risk" alarm, reminding the doctor to consider adjusting the treatment plan, such as increasing anti-inflammatory treatment, optimizing diabetes control strategies, and increasing rehabilitation training.

[0258] Under the traditional assessment system, doctors would only realize the progression of Mr. Li's weakness when his muscle mass significantly decreased and his walking speed dropped to 9.5s / 6m. However, the system of this invention provides an early warning 3.2 months in advance, enabling doctors to take intervention measures earlier.

[0259] To verify the effectiveness of the method of the present invention, we collected data from 1,200 elderly cancer patients between 2021 and 2023, of which 600 were used for training, 300 for testing, and 300 for clinical validation, and compared them with traditional methods.

[0260] method Prediction accuracy Early warning time Key feature number Feature redundancy rate Short-term forecast error Long-term forecast error Method of the present invention 91.4% 3.2 months in advance 10 15% 12.8% 17.3% Traditional Fried exponent method 72.3% No advance warning 5 40% 22.5% 34.1% Traditional Rockwood rating method 75.1% No advance warning 6 38% 20.2% 30.7%

[0261] The results show that the method of this invention improves the prediction accuracy by 16%-19% compared to traditional methods and can identify the risk of frailty 3.2 months in advance. Furthermore, by optimizing feature selection, this invention reduces feature redundancy (from 40% to 15%), improving the computational efficiency and medical interpretability of the model. Simultaneously, in terms of short-term (1 month) and long-term (6 months) health status prediction errors, the method of this invention reduces them by approximately 10% and 13% compared to traditional methods, respectively, significantly improving the reliability of predicting early-stage frailty risks.

[0262] In this case, the frailty risk prediction system for elderly cancer patients of the present invention successfully identified Mr. Li's frailty risk 3.2 months in advance and triggered a high-risk alert before September 15, allowing doctors to adjust the treatment plan in time. Experimental data show that the system is significantly better than traditional methods in terms of prediction accuracy, early warning capability, feature selection accuracy and model stability, providing a more scientific and accurate intelligent prediction tool for the health management of elderly cancer patients.

[0263] This invention employs an improved discrete monarch butterfly optimization algorithm for feature selection. By combining a global search strategy with the experience of medical experts, it effectively optimizes the key feature subset required for predicting the pre-frailty risk of elderly cancer patients and introduces a hybrid discrete migration strategy. In the feature subset optimization process, individual migration and local perturbation operators are combined to ensure that feature selection takes into account both global search capability and local optimization accuracy, avoids feature redundancy, and reduces computational overhead.

[0264] In the prediction process, this invention innovatively combines a multi-scale Markov chain model to model the health status of elderly cancer patients at different time scales, making up for the shortcomings of traditional static prediction models that are difficult to capture the trend of changes in patients' health status. Furthermore, by using short-term, medium-term, and long-term health status transition probability matrices, a hierarchical prediction mechanism is established, enabling the model to conduct a comprehensive analysis from short-term health fluctuations and medium-term trends to long-term risk evolution.

[0265] In the risk assessment stage, this invention employs a multi-scale dynamic risk prediction method. Based on the risk status sequence of elderly cancer patients, it integrates short-term, medium-term, and long-term prediction information to construct a global risk trend analysis model. Combined with preset risk warning thresholds, the warning level is dynamically adjusted to predict the risk of future frailty in the early stages of changes in the patient's health status, providing the possibility of early intervention. In addition, this invention uses dynamic risk change rate calculation and combines risk warning levels to accurately trigger personalized risk warnings, avoiding false alarms or missed alarms caused by oversensitivity or lag in traditional methods.

[0266] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A pre-frailty risk prediction system for elderly cancer patients based on an improved population algorithm, characterized in that, include: The data acquisition module collects multidimensional data from elderly cancer patients and constructs a dataset of elderly cancer patients. The data preprocessing module preprocesses the dataset of elderly cancer patients to generate a fused feature matrix of elderly cancer patients. The feature processing module, based on the feature matrix of elderly cancer patients, uses statistical analysis and medical expert experience to perform preliminary screening of various features in the data of elderly cancer patients, forming an initial feature set; The feature selection module inputs the initial feature set into the discrete monarch butterfly optimization algorithm, and uses a global search strategy to optimize and select features from the data of elderly cancer patients, generating a subset of key features; The multi-scale Markov chain modeling module constructs a multi-scale Markov chain model based on the key feature subset, models the health status of elderly cancer patients at different time scales, defines the patient state and state transition probability matrix, and forms a risk state sequence. The dynamic risk prediction module dynamically predicts the pre-frailty risk of elderly cancer patients based on the risk state sequence and obtains real-time risk assessment results. It compares the real-time risk assessment results with the preset risk warning threshold to determine whether the pre-frailty risk of elderly cancer patients reaches the warning standard and generates corresponding risk warning information. The multi-scale Markov chain modeling module includes: Based on the aforementioned key feature subset Extract observational data of elderly cancer patients at different time points to construct a time series of their health status: in, For the key feature matrix at each time point Observations at that location ,in Indicates the first At the first time point Observations of key features; Discretize the pre-frailty health status of elderly cancer patients, mapping the patient's status to a finite set of health levels: in, A set of health states, representing different health states, including normal. Mild weakness Moderate weakness and severe weakness ; A threshold segmentation method is used based on the statistical distribution of key feature subsets to segment the observations at each time point. Mapped to a set of health states : in For the first Health status at a specific point in time Representing observation data Belonging to state The probability of; Establishing multi-scale Markov chain models at different time scales : in, Representing the time scale, establishing short-term... Mid-term and long-term Markov chain model, The state transition probability matrix represents the changing pattern of health status at different time scales. Multiscale Markov chain model In the short timescale Next, calculate the transition probability of health status: in, The probability of a short-term health status transition indicates the patient's transition from a certain state to a certain condition. Change to state The probability, and These represent the health status at adjacent time points, and the short-term transition probability is estimated based on follow-up data. Multiscale Markov chain model In the medium timescale The following calculation is performed using the state accumulation transition method: in, The probability of transitioning to a new health status in the intermediate stage represents the patient's transition from a certain state. Change to state The probability, It reflects the probability of two-step state transitions and estimates the long-term trend through intermediate transitions between states; Multiscale Markov chain model On a long time scale Below, the long-term trend is calculated using steady-state probability: in, The probability of transitioning to a long-term healthy state represents the patient's transition from a certain state. Change to state The probability, It reflects the evolution trend of long-term health status and predicts the long-term health risk of elderly cancer patients by calculating the steady-state distribution of the transfer matrix. The calculated state transition probability matrices at different time scales are used to calculate the patient's risk state sequences in the short, medium, and long term: in, Indicated on the time scale Below, the patient is Predicting health status in real time To allow patients to transition from their current healthy state over a timescale d. Changes to future health status The state transition probability.

2. A method for predicting the pre-frailty risk of elderly cancer patients based on an improved population algorithm, applied to the pre-frailty risk prediction system for elderly cancer patients based on an improved population algorithm as described in claim 1, characterized in that, Includes the following steps: S1. Collect physiological parameter data, pathology report data, electronic medical record data, and physician evaluation data from elderly cancer patients to construct a dataset of elderly cancer patients; S2. Preprocess the dataset of elderly cancer patients to generate a fused feature matrix of elderly cancer patients; S3. Based on the feature matrix of elderly cancer patients, statistical analysis and medical expert experience are used to initially screen various features in the data of elderly cancer patients to form an initial feature set; S4. Input the initial feature set into the discrete monarch butterfly optimization algorithm, and use a global search strategy to optimize and select features in the data of elderly cancer patients to generate a subset of key features; S5. Construct a multi-scale Markov chain model based on the aforementioned key feature subset, model the health status of elderly cancer patients at different time scales, define patient status and state transition probability matrix, and form a risk state sequence; S6. Dynamically predict the pre-frailty risk of elderly cancer patients based on the risk status sequence, obtain real-time risk assessment results, compare the real-time risk assessment results with the preset risk warning threshold, determine whether the pre-frailty risk of elderly cancer patients reaches the warning standard, and generate corresponding risk warning information.

3. The method for predicting pre-frailty risk in elderly cancer patients based on an improved population algorithm according to claim 2, characterized in that, S1 includes the following steps: S11. Collect physiological parameter data from elderly cancer patients. This includes heart rate, blood pressure, blood oxygen saturation, body temperature, and blood glucose levels; S12. Collect pathology report data from elderly cancer patients. This includes tumor histological type, pathological stage, and expression levels of related biomarkers; S13. Collect electronic medical record data of elderly cancer patients This includes the patient's basic information, past medical history, family medical history, treatment plan, and medication details; S14. Collect physician assessment data for elderly cancer patients This includes subjective assessment scores, frailty scores, functional status scores, and cognitive ability assessments; S15. Based on the timestamp information, the collected physiological parameter data, pathology report data, electronic medical record data, and physician evaluation data of elderly cancer patients are time-aligned to construct a dataset of elderly cancer patients: ； in, This is the final dataset of elderly cancer patients.

4. The method for predicting pre-frailty risk in elderly cancer patients based on an improved population algorithm according to claim 2, characterized in that, S2 includes the following steps: S21. Dataset of elderly cancer patients Perform data format conversion to transform data from different sources into a unified storage format and construct a standardized dataset of elderly cancer patients; S22. Perform data cleaning on the standardized elderly cancer patient dataset to remove redundant and duplicate data, forming a cleaned elderly cancer patient dataset. S23. Detect outliers in the cleaned elderly cancer patient dataset, and use statistical methods to process the outliers to generate an outlier-processed elderly cancer patient dataset. S24. After outlier processing, detect missing data in the elderly cancer patient dataset, and use mean imputation, interpolation imputation, or imputation based on historical data to process the missing data, and obtain the elderly cancer patient dataset after missing data imputation. S25. After imputing the missing data, the dataset of elderly cancer patients is normalized so that each feature value is mapped to the same numerical range, thus obtaining the normalized dataset of elderly cancer patients. S26. Standardize the normalized dataset of elderly cancer patients so that the data features conform to a standard normal distribution, thus obtaining the standardized dataset of elderly cancer patients. And generate a fused feature matrix of elderly cancer patients. .

5. A method for predicting pre-frailty risk in elderly cancer patients based on an improved population algorithm, as described in claim 2, is characterized in that... S3 includes the following steps: S31. From the fused feature matrix of elderly cancer patients Extract all feature variables and construct a feature set for elderly cancer patients: ； in, This is a set of characteristics of elderly cancer patients. Indicates the first Feature variables, The total number of features; S32. Calculate the feature set of elderly cancer patients The mean of each feature variable and standard deviation Features with variability below a threshold in all patient samples were removed to form a feature set of elderly cancer patients after variability screening. ; S33. Calculate the feature set of elderly cancer patients after variability screening. The correlation between various characteristic variables and the frailty status of elderly cancer patients was analyzed. Characteristic variables with correlations below a set threshold were removed. Based on the experience of medical experts, the characteristic set of elderly cancer patients after correlation screening was further refined. A clinical significance assessment was conducted, and medically insignificant characteristic variables were removed to form a characteristic set of elderly cancer patients after medical screening. ； in, A set of characteristics of elderly cancer patients after medical screening; S34. Calculate the characteristic set of elderly cancer patients after medical screening based on statistical analysis methods. The importance weights of each feature variable are determined, and the top-weighted variables are selected. The characteristic variables ultimately form the initial feature set: ； in, This is the initial set of features selected for the final screening.

6. A method for predicting pre-frailty risk in elderly cancer patients based on an improved population algorithm, as described in claim 2, is characterized in that... S4 includes the following steps: S41. Based on the initial feature set Constructing an initial discrete monarch butterfly population matrix for predicting pre-frailty risk in elderly cancer patients. : in, For the first The initial feature selection vector of an individual monarch butterfly. The first indicator of pre-frailty risk in elderly cancer patients This indicates whether a feature is selected; 1 means selected, and 0 means not selected. , For population size, The total number of features in the initial feature set; The initialization of the discrete monarch butterfly population was based on the statistical distribution of clinical data, which enabled it to cover the characteristic patterns of pre-debilitative elderly cancer patients of different types in its initial state. S42. Based on the accuracy, feature importance, and data redundancy of predicting pre-frailty risk in elderly cancer patients, a fitness function is constructed: ； in, Indicates based on the first The accuracy of the pre-frailty risk prediction model for elderly cancer patients constructed from a subset of features. This represents the dimension of the selected feature subset. Representing characteristic variables and The correlation between them These are the weighting coefficients; Calculate the initial fitness value of the fitness function. It is used to sort the population individuals and determine the initial global optimum. ; S43. Calculate the fitness value of all individuals in the discrete monarch butterfly population, and determine the current global optimal solution based on the fitness value. Then, the feature selection vectors of the migrating subgroups are updated, and a hybrid discrete migration strategy is adopted to bring them closer to the optimal solution: ； in, This represents the selection state of the i-th feature of the j-th migrating subgroup individual after iteration t+1. This represents the selection state of the globally optimal feature subset on the i-th feature at iteration t. and They are two independent random numbers, For adaptive migration probability, and fitness value Positive correlation This represents the selection state of the i-th feature for the j-th individual in the current migratory subgroup during the t-th iteration; S44. For non-migratory subgroups, feature subset optimization is performed using a local perturbation operator guided by medical prior knowledge: ； in, For local perturbation probability, and Related, For medical experts to identify characteristics Clinical importance assessment value, S45. Fitness values ​​of fused migratory subpopulations and non-migratory subpopulations: in, The fusion weight depends on the mean fitness of the migrating and non-migrating subgroups. Select vectors for the individual features of monarch butterflies updated after the (t+1)th iteration; Calculate the new fitness: And update the globally optimal individual: in, Let represent the set of fitness values ​​of all monarch butterfly individuals after the (t+1)th iteration. This represents the monarch butterfly individual with the highest fitness after the (t+1)th iteration, which is the globally optimal feature subset of the current iteration. S46. The Discrete Monarch Butterfly optimization algorithm stops when any of the following conditions are met: Reaching the maximum number of iterations ; Continuous T rounds of optimization, That is, the globally optimal individual no longer changes, and eventually converges to obtain the globally optimal individual: S47. The final globally optimal individual Corresponding key feature subset : 。 7. A method for predicting pre-frailty risk in elderly cancer patients based on an improved population algorithm, as described in claim 2, is characterized in that... S6 includes the following steps: S61. Based on patient risk status sequence To construct an individualized pre-frailty risk probability for elderly cancer patients: in, Indicated on the time scale Under these circumstances, elderly cancer patients Constantly in the early stages of weakness or more severe conditions, i.e., mild weakness. Moderate weakness and severe weakness The probability of; S62. Based on the multi-scale pre-weakness risk probability, and integrating short-term, medium-term, and long-term risks, calculate the global risk trend: in, Indicates in At any given moment, the probability of early-stage weakness is determined by combining short-term, medium-term, and long-term forecasts. Weights for different time scales; S63. Based on the overall risk trend, assess the risk change trend in the early stages of decline: in, Indicates in At any given time, the patient's pre-deterior risk rate changes; if This indicates an increased risk of deterioration in the patient. This indicates a reduced risk for the patient; S64. Pre-set a set of risk warning thresholds: in, This indicates a low-risk threshold, corresponding to a mild risk of weakening. This represents the medium risk threshold, corresponding to a moderate risk of weakening. This indicates a high-risk threshold, corresponding to a severe weakening risk; S65. Assess the global risk probability based on the risk warning threshold. Risk level assessment: ； in, Indicates in Risk level at any given moment; S66. Based on the rate of change of risk Risk level determination Generate personalized dynamic risk warnings: ； in, Indicates in The system generates dynamic risk warning information in real time. If the patient's risk level is below the threshold and there is no obvious upward trend, the system will not trigger a warning. If the risk level is moderate or the risk is rising, a warning reminder will be triggered. If the risk level is high or the rate of increase exceeds the high-risk threshold, a high-risk alarm will be triggered.

Images