Postoperative sleep disorder key factor identification method

Abstract

The invention discloses a postoperative sleep disorder key factor identification method, and belongs to the technical field of medical informatics, and the method comprises the steps: collecting data, and preprocessing the data into a candidate feature set; carrying out feature importance calculation by adopting evaluation methods of at least five different principles; the importance and stability of the features are evaluated through score and ranking analysis path fusion; the features are divided into four quadrants according to importance and stability indexes by using a two-dimensional scatter diagram, and feature classification is performed, so that key factors of the postoperative sleep disorder are accurately positioned. The method improves the robustness and consistency of feature evaluation through multi-model fusion, reduces the deviation of a single method, achieves the precise recognition of key factors of postoperative sleep disorders, supports the construction and intervention optimization of a clinical prediction model, and improves the diagnosis accuracy and patient management efficiency.

Description

Technical Field

[0001] This invention belongs to the field of medical informatics technology, specifically relating to a method for identifying key factors of postoperative sleep disorders. Background Technology

[0002] Postoperative sleep disorder (POSD) is a common clinical complication, characterized by disrupted sleep structure, decreased efficiency, difficulty falling asleep, and frequent awakenings. It not only affects patient comfort but is also associated with adverse outcomes such as delayed wound healing, increased risk of infection, and postoperative delirium, thus prolonging hospital stays and increasing costs. Traditional empirical or univariate statistical methods struggle to handle high-dimensional, multi-source clinical data, limiting the identification of key factors. With the widespread adoption of electronic medical records, machine learning techniques (such as random forests, GBDT, and SVM) have been introduced into risk prediction, providing new avenues for identifying key factors when combined with feature importance assessments (Gini importance, permutation importance, and SHAP values). However, single models or methods are biased; different algorithms often yield contradictory factor rankings due to mechanistic differences, severely impacting the stability and transferability of conclusions. Although some studies have attempted to mitigate discrepancies through fusion strategies such as average ranking, these methods ignore ranking consistency information, fail to distinguish between "stable and consistent" and "highly controversial" features, and lack standardized tools for quantifying the stability of individual features. This deficiency makes it difficult for clinical researchers to determine which factors are truly reliable when faced with conflicting conclusions, thus hindering the extraction and application of precise factors. Therefore, there is an urgent need for an innovative method that can systematically integrate the importance results of multiple models and methods, and quantify and visualize the overall importance and stability of each feature. This would provide clinicians with a more robust and interpretable factor identification framework, promoting the application and development of precision medicine in postoperative rehabilitation. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method for identifying key factors of postoperative sleep disorders by improving the robustness and consistency of feature assessment through multi-model fusion, reducing the bias of single methods, achieving accurate identification of key factors of postoperative sleep disorders, supporting the construction of clinical prediction models and intervention optimization, and improving diagnostic accuracy and patient management efficiency.

[0004] The technical solution of this invention is as follows:

[0005] A method for identifying key factors of postoperative sleep disorders includes the following steps: Step 1: Data preparation and preprocessing, including data acquisition and data preprocessing. Data acquisition involves obtaining multi-dimensional patient data from electronic medical record systems, monitoring devices, or clinical databases to form a candidate feature set. , where n is the total number of features, and whether or not sleep disorder occurs after surgery is determined as a binary classification label; data preprocessing cleans and transforms the raw data, including missing value imputation, outlier handling, data type conversion, and classification feature encoding; Step 2: Multi-source feature importance generation, including method selection, model training details, and importance calculation. Method selection involves pre-selecting at least five feature importance evaluation methods with different principles. Model training details include data splitting, hyperparameter tuning, handling class imbalance, and model evaluation. Importance calculation uses the selected methods to analyze the preprocessed data and calculate importance for each feature. Calculate importance, and output the result as a score or ranking; Step 3: Feature importance and stability fusion assessment, providing score-based and ranking-based analysis paths. The score-based analysis path includes score normalization and fusion index calculation, calculating the average importance score and feature consistency index; the ranking-based analysis path includes ranking acquisition and fusion index calculation, calculating the average ranking and feature importance stability index. Step 4: Four-quadrant analysis and key factor classification, including two-dimensional visualization and quadrant division and feature classification. The two-dimensional visualization constructs a scatter plot with importance index as the X-axis and stability index as the Y-axis. The quadrant division divides the plot into four quadrants according to the threshold and classifies the features.

[0006] Furthermore, in the data collection of step one, the multidimensional data includes demographic characteristics, preoperative comorbidity characteristics, surgery and anesthesia-related characteristics, and postoperative management characteristics; the binary labels are determined by the Pittsburgh Sleep Quality Index or polysomnography.

[0007] Furthermore, in the data preprocessing of step one, the missing value imputation adopts the mean, median or model prediction method; the classification feature encoding adopts one-hot encoding or target encoding.

[0008] Furthermore, in the method selection in step two, the five methods include global importance based on SHAP, importance based on permutation, built-in gain importance based on gradient boosting decision tree model, local importance based on LIME, and coefficient importance based on Lasso regression.

[0009] Furthermore, in the model training details of step two, the data segmentation uses k-fold cross-validation; the hyperparameter tuning sets the learning rate, maximum depth, and number of iterations for the specific model; the handling of class imbalance uses oversampling or weight adjustment; and the model evaluation uses accuracy, precision, and recall.

[0010] Furthermore, in the score-based analysis path of step three, the score is normalized as follows: for each method The resulting importance score vector Perform min-max normalization on all scores. Mapping to the [0,1] interval yields the normalized score. The formula is as follows: , In the formula, The normalized score represents the characteristic. In the method The standardized importance value is given below, with a range of [0,1], where 0 represents the lowest importance and 1 represents the highest importance; The original importance scores represent the method. Features The initial evaluation score is directly output by the importance evaluation algorithm, reflecting the relative contribution of the feature in the model; For method The generated importance score vector of all features represents the method. For the entire feature set The set of fractions; The fusion index is calculated as follows: for each feature Based on its m normalized scores Calculate the following two core metrics: (1) Average Importance Score: This indicator measures the average importance level of a feature, and the formula is as follows: , Its range is [0,1], and the larger the value, the higher the overall importance of the feature under multiple methods; (2) Feature Consistency Index: This index measures the stability and consistency of the conclusions regarding feature importance. The formula is as follows: , In the formula, Features of The sample variance of the normalized scores. The value of FCI is 0.25, which is the theoretical maximum variance of the variable in the interval [0,1], and is used to ensure that the value range of FCI falls within the interval [0,1]. The closer the value of FCI is to 1, the more consistent the different methods are in assessing the importance of this feature, and the more stable the conclusion is. The closer the value of FCI is to 0, the greater the divergence, and the less stable the conclusion is.

[0011] Furthermore, in the ranking-based analysis path of step three, the ranking is obtained as follows: for each method This yields a feature ranking list. , where the i-th feature The ranking is recorded as ; The fusion index is calculated as follows: for each feature Based on its Ranking Calculate the following two indicators: (1) Average ranking, calculated using the following formula: , The smaller the average ranking value, the higher the overall importance. (2) Feature importance stability index, calculated using the following formula: , In the formula, As a feature of The sample variance of each ranking. ,yes The theoretical maximum variance of the feature ranking [0,1] is used for normalization; The closer the FISI value is to 1, the more stable the ranking.

[0012] Furthermore, in the quadrant division and feature classification in step four, the threshold is defined by experts in the field or adaptively determined based on the data distribution, including the use of the median or quartiles; the feature classification is to classify features falling into different quadrants, namely: first quadrant: core key factors; second quadrant: controversial factors; third quadrant: stable irrelevant factors; fourth quadrant: noise factors.

[0013] The beneficial effects of this invention are: 1. Improved reliability and robustness of key factor identification: This invention effectively overcomes the inherent biases and limitations of relying on a single model or method by systematically integrating multiple evaluation methods based on different principles (such as SHAP, permutation importance, and model built-in gain). This multi-source information fusion mechanism can smooth out anomalous results from individual methods, making the final identified key factor conclusions more stable and reliable. It greatly reduces the risk of biased research conclusions due to inappropriate method selection, and the results have stronger reproducibility and credibility.

[0014] 2. Innovatively, this invention achieves a quantitative assessment of the stability of feature importance conclusions: For the first time, this invention introduces the core concept of Feature Consistency Index (FCI) or Feature Importance Stability Index (FISI), filling the gap in existing technologies that lack quantitative assessments of the consistency of individual features. Unlike existing simple average ranking methods, this invention not only focuses on the "average importance" of features but also reveals the "consensus" of the importance conclusion. This allows researchers to clearly distinguish whether a feature is "generally recognized as important" or "controversial," providing a quantitative basis for assessing the credibility of conclusions.

[0015] 3. Provides an actionable decision-making framework: The "four-quadrant analysis method" of this invention combines the two dimensions of "importance" and "stability" of features, providing a precise classification for all candidate features. By dividing features into "core key factors," "controversial factors," "stable irrelevant factors," and "noise factors," this method not only indicates which features are most important but also reveals their behavioral patterns. In particular, the identification of "controversial factors" can guide researchers to conduct more in-depth exploratory analysis, discover potential complex interactions, thereby greatly improving R&D efficiency and research depth, and providing a decision-making framework that is far more insightful than a single ranking list. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the process steps of a method for identifying key factors of postoperative sleep disorders according to the present invention. Detailed Implementation

[0017] like Figure 1 As shown, a method for identifying key factors of postoperative sleep disorders includes the following steps: Step 1: Data preparation and preprocessing, including data acquisition and data preprocessing. Data acquisition involves obtaining multi-dimensional patient data from electronic medical record systems, monitoring devices, or clinical databases to form a candidate feature set. , where n is the total number of features, and whether or not sleep disorder occurs after surgery is determined as a binary classification label; data preprocessing cleans and transforms the raw data, including missing value imputation, outlier handling, data type conversion, and classification feature encoding; Step 2: Multi-source feature importance generation, including method selection, model training details, and importance calculation. Method selection involves pre-selecting at least five feature importance evaluation methods with different principles. Model training details include data splitting, hyperparameter tuning, handling class imbalance, and model evaluation. Importance calculation uses the selected methods to analyze the preprocessed data and calculate importance for each feature. Calculate importance, and output the result as a score or ranking; Step 3: Feature importance and stability fusion assessment, providing score-based and ranking-based analysis paths. The score-based analysis path includes score normalization and fusion index calculation, calculating the average importance score and feature consistency index; the ranking-based analysis path includes ranking acquisition and fusion index calculation, calculating the average ranking and feature importance stability index. Step 4: Four-quadrant analysis and key factor classification, including two-dimensional visualization and quadrant division and feature classification. The two-dimensional visualization constructs a scatter plot with importance index as the X-axis and stability index as the Y-axis. The quadrant division divides the plot into four quadrants according to the threshold and classifies the features.

[0018] Preferably, in the data collection of step one, the multidimensional data includes demographic characteristics, preoperative comorbidity characteristics, surgery and anesthesia-related characteristics, and postoperative management characteristics; the binary labels are determined by the Pittsburgh Sleep Quality Index or polysomnography.

[0019] Preferably, in the data preparation process of step one, data acquisition obtains multi-dimensional patient data from electronic medical record systems, monitoring devices, or clinical databases to form a candidate feature set. This multi-dimensional data specifically includes three categories of indicators: preoperative, intraoperative, and postoperative. Each category of indicators has a potential correlation with the occurrence and development of postoperative sleep disorders. Preoperative factors: age, body mass index (BMI), presence of diabetes, presence of hypertension, presence of at least two comorbidities, surgical history, American Society of Anesthesiologists Health Classification (ASA), preoperative laboratory test data: white blood cell count, albumin, electrolytes, electrocardiogram, presence of anemia, cardiac function classification, and Self-Rating Anxiety Scale (SAS) score.

[0020] Intraoperative factors: type of surgery, start time of surgery, duration of surgery, amount of crystalloid solution used during surgery, whether colloid solution was infused, blood transfusion, amount of blood loss, medication used (dexamethasone, midazolam, sufentanil, propofol, remifentanil, dexmedetomidine), whether hypotension occurred, duration of anesthesia, and duration of recovery.

[0021] Postoperative factors: whether there is anemia after surgery, postoperative white blood cell count, postoperative albumin, postoperative electrolytes, whether there is hypothermia, postoperative pain score NRS, postoperative nausea and vomiting, postoperative respiratory status, postoperative digestive status, whether there is sore throat, whether there is dizziness, and whether a pain pump is used.

[0022] The above indicators cover comprehensive information about patients during the perioperative period, and their relationship with postoperative sleep disorders can be clarified through factor classification results: Key factors include: surgery duration, postoperative use of a pain pump, and anesthesia duration. These factors are closely related to postoperative sleep disturbances and are the main drivers affecting postoperative sleep quality.

[0023] Controversial factors include: preoperative sleep status, American Society of Anesthesiologists Health Classification (ASA), preoperative white blood cell count, preoperative electrolytes, preoperative electrocardiogram, cardiac function classification, intraoperative crystalloid fluid dosage, whether colloid fluids were infused, blood transfusion, blood loss, whether dexamethasone was used, propofol dosage, whether dexmedetomidine was used intraoperatively, whether intraoperative hypotension occurred, recovery time, postoperative pain score (NRS), postoperative nausea and vomiting, postoperative respiratory status, postoperative digestive status, postoperative sore throat, and dizziness. These factors are of high importance, but different methods of assessment yield discrepancies, requiring further research.

[0024] Stable irrelevant factors: remifentanil dosage, postoperative dizziness, sufentanil dosage, preoperative albumin, surgical history, recovery time, preoperative white blood cell count, type of surgery, time of surgery start, use of midazolam, presence of hypertension, presence of diabetes, and presence of at least two comorbidities. These factors were consistently considered unimportant in multiple methods.

[0025] Noise factors: Self-Rating Anxiety Scale (SAS) score, propofol dosage, age, and body mass index (BMI). These factors have low importance and inconsistent assessments, and may be confounding items in the data.

[0026] In addition, the Pittsburgh Sleep Quality Index (PSQI) or polysomnography was used to determine whether patients experienced postoperative sleep disorders, which served as a binary label; 0 for no sleep disorders and 1 for sleep disorders.

[0027] In the data preprocessing of step one, the missing value imputation uses the mean, median, or model prediction method; the classification feature encoding uses one-hot encoding or target encoding. Outlier handling refers to identifying and managing values ​​that significantly deviate from the normal distribution or expected range of the dataset during the data preprocessing stage. These outliers may originate from data entry errors, measurement biases, or real-world extreme events. If left unhandled, they may lead to bias or overfitting in machine learning models. In this method, outlier handling is part of the cleaning step, aiming to improve data quality. Specifically, it includes the following key operations: Outlier detection: Statistical methods are used to identify outliers, such as interquartile range (IQR) based methods, where outliers are defined as values ​​below Q1 - 1.5 × IQR or above Q3 + 1.5 × IQR (Q1 and Q3 are the first and third quartiles, respectively); or Z-score methods, where values ​​with Z-scores exceeding a threshold (e.g., 3) are considered outliers. Handling strategies: Once outliers are detected, removal (deleting outlier records, applicable to erroneous data), replacement (filling with the mean, median, or pattern, applicable to continuous variables), truncation (Winsorization, adjusting outliers to threshold boundaries), or retention and labeling (if the outlier is clinically significant, such as extreme pain scores) can be employed. In the context of postoperative sleep disorder datasets, for example, if the BMI value is abnormally high (e.g., exceeding 100), it can be detected by IQR and replaced with the median to avoid affecting model training. The data type conversion involves adjusting the format of the raw data from one type to another to ensure compatibility and computational efficiency with machine learning algorithms. This process is a core component of preprocessing because many models require inputs of a specific type (such as numerical). In this method, data type conversion is optimized for the diversity of clinical data (such as numerical, categorical, and date data), specifically including: Numeric type conversion: For example, converting a string representing a numerical value (such as "65" representing age) to an integer or floating-point number; or rounding a floating-point number to an integer if high precision is not required; Date and time conversion: Convert string dates (such as "2023-01-01") into datetime objects to facilitate the calculation of time-related features such as surgery duration; Category type conversion: Although related to encoding, conversion can include converting categorical strings (such as "male / female") into categorical types to reduce memory usage and speed up processing. In postoperative sleep disorder data, for example, converting surgery type (such as "open surgery / minimally invasive") from a string into a category label. This step is typically implemented using programming libraries (such as Python's Pandas), for example, through the astype() function, to ensure that all features meet the model input requirements, thereby supporting subsequent multi-source feature importance generation.

[0028] Preferably, in the method selection in step two, the five methods include global importance based on SHAP, importance based on permutation, built-in gain importance based on gradient boosting decision tree model, local importance based on LIME, and coefficient importance based on Lasso regression. The specific implementation method is as follows: each method independently calculates feature importance on the preprocessed dataset, and the results are used for subsequent fusion evaluation: (1) Global Importance Based on SHAP: The SHAP (SHapley Additive exPlanations) method evaluates global importance by calculating the contribution of each feature to the model's prediction. In specific implementations, TreeExplainer (suitable for tree-based models) or KernelExplainer (suitable for any model) from the SHAP library are used. For example, after training an XGBoost model, the SHAP value is calculated as follows: import shap explainer = shap.TreeExplainer(model) # model is the trained XGBoost model shap_values ​​= explainer.shap_values(X) # X is the feature matrix global_importance = np.abs(shap_values).mean(axis=0) # Global importance is the average of absolute SHAP values; This method outputs the SHAP score for each feature, reflecting its impact on the overall prediction; (2) Permutation-based importance: Permutation importance is measured by randomly permuting feature values ​​and observing the decrease in model performance. Specifically, the permutation_importance function of sklearn is used. For example, for the random forest model: from sklearn.inspection import permutation_importance result = permutation_importance(model, X_val, y_val, n_repeats=10, random_state=42) # X_val, y_val are the validation sets importance = result.importances_mean # Average performance degradation value is used as the importance score; This method outputs a score vector; (3) Built-in gain importance of gradient boosting decision tree model: Using the built-in feature importance of XGBoost or LightGBM, the gain is calculated based on the tree split. Specific implementation: import xgboost as xgb model = xgb.XGBClassifier() # Train the model model.fit(X_train, y_train) gain_importance = model.get_booster().get_score(importance_type='gain') # Gain importance; This method outputs the score in dictionary form; (4) Local importance based on LIME: LIME (Local Interpretable Model-agnostic Explanations) evaluates importance by approximating global behavior through local linear models. Specifically, it uses the LIME library to aggregate the local interpretations of multiple instances to obtain a global view. import lime_tabular from lime explainer = lime_tabular.LimeTabularExplainer(X_train, feature_names=feature_names, class_names=['No POSD', 'POSD']) local_exps = [explainer.explain_instance(X_test[i], model.predict_proba) for i in range(len(X_test))] global_importance = np.mean([np.abs(exp.as_map()[1]) for exp inlocal_exps], axis=0) # Aggregate absolute coefficients; This method outputs the average coefficient; (5) Coefficient importance based on Lasso regression: Lasso achieves feature selection by shrinking the coefficient through L1 regularization. Specifically, the Lasso model from sklearn is used. from sklearn.linear_model import Lasso model = Lasso(alpha=0.1) # alpha is the regularization strength model.fit(X_train, y_train) coef_importance = np.abs(model.coef_) # Absolute coefficients are used as importance; This method outputs a coefficient vector; The selection of the above five methods ensures the diversity of evaluation perspectives, and the output results can be scores or rankings, which can be used for integration in step three.

[0029] Preferably, in the model training details of step two, the data segmentation uses k-fold cross-validation; the hyperparameter tuning sets the learning rate, maximum depth, and number of iterations for the specific model; the handling of class imbalance uses oversampling or weight adjustment; and the model evaluation uses accuracy, precision, and recall.

[0030] In this embodiment, the training process of the underlying model (such as XGBoost for SHAP and gain importance, and random forest for permutation importance) is described in detail. The dataset is divided into a training set (70%) and a validation set (30%) to handle the imbalance of POSD labels (e.g., positive samples account for 30%).

[0031] (1) Data partitioning: k-fold cross-validation (k=5) is used to evaluate model stability. Specific implementation: from sklearn.model_selection import StratifiedKFold skf = StratifiedKFold(n_splits=5, shuffle=True, random_state=42) for train_index, val_index in skf.split(X, y): X_train, X_val = X[train_index], X[val_index] y_train, y_val = y[train_index], y[val_index] # Train the model on each fold; This approach ensures a balanced distribution across each compromise category; (2) Hyperparameter tuning: Set key hyperparameters for the specific model. For example, for XGBoost: learning rate (learning_rate=0.1), maximum depth (max_depth=6), and number of iterations (n_estimators=100), and optimize using grid search. from sklearn.model_selection import GridSearchCV param_grid = {'learning_rate': [0.01, 0.1], 'max_depth': [4, 6], 'n_estimators': [50, 100]} grid_search = GridSearchCV(model, param_grid, cv=3, scoring='roc_auc') grid_search.fit(X_train, y_train) best_params = grid_search.best_params_ # The best parameters are used in the final model; (3) Handling class imbalance: Use oversampling (SMOTE) or weight adjustment. Specifically, implement SMOTE as follows: from imblearn.over_sampling import SMOTE smote = SMOTE(random_state=42) X_resampled, y_resampled = smote.fit_resample(X_train, y_train) # Train after balancing; Or weight adjustment: model = xgb.XGBClassifier(scale_pos_weight= (len(y_train) - sum(y_train)) / sum(y_train)) # Positive sample weights; (4) Model Evaluation: The model performance is evaluated using accuracy, precision, and recall. Specific implementation details are as follows: from sklearn.metrics import accuracy_score, precision_score, recall_score y_pred = model.predict(X_val) acc = accuracy_score(y_val, y_pred) # Accuracy prec = precision_score(y_val, y_pred) # Precision rate rec = recall_score(y_val, y_pred) # Recall rate; These training details improve the model's generalization ability and fairness, and support the accurate generation of the importance of multi-source features.

[0032] Preferably, in the score-based analysis path of step three, the score is normalized as follows: for each method The resulting importance score vector Perform min-max normalization on all scores. Mapping to the [0,1] interval yields the normalized score. The formula is as follows: , In the formula, The normalized score represents the characteristic. In the method The standardized importance value is given below, with a range of [0,1], where 0 represents the lowest importance and 1 represents the highest importance; The original importance scores represent the method. Features The initial evaluation score is directly output by the importance evaluation algorithm, reflecting the relative contribution of the feature in the model; For method The generated importance score vector of all features represents the method. For the entire feature set The set of fractions; The fusion index is calculated as follows: for each feature Based on its m normalized scores Calculate the following two core metrics: (1) Average Importance Score: This indicator measures the average importance level of a feature, and the formula is as follows: , Its range is [0,1], and the larger the value, the higher the overall importance of the feature under multiple methods; (2) Feature Consistency Index: This index measures the stability and consistency of the conclusions regarding feature importance. The formula is as follows: , In the formula, Features of The sample variance of the normalized scores. The value of FCI is 0.25, which is the theoretical maximum variance of the variable in the interval [0,1], and is used to ensure that the value range of FCI falls within the interval [0,1]. The closer the value of FCI is to 1, the more consistent the different methods are in assessing the importance of this feature, and the more stable the conclusion is. The closer the value of FCI is to 0, the greater the divergence, and the less stable the conclusion is.

[0033] Preferably, in the ranking-based analysis path of step three, the ranking is obtained as follows: for each method This yields a feature ranking list. , where the i-th feature The ranking is recorded as ; The fusion index is calculated as follows: for each feature Based on its Ranking Calculate the following two indicators: (1) Average ranking, calculated using the following formula: , The smaller the average ranking value, the higher the overall importance. (2) Feature importance stability index, calculated using the following formula: , In the formula, As a feature of The sample variance of each ranking. ,yes The theoretical maximum variance of the feature ranking [0,1] is used for normalization; The closer the FISI value is to 1, the more stable the ranking.

[0034] Preferably, in the quadrant division and feature classification in step four, the threshold is defined by experts in the field or adaptively determined based on the data distribution, including the use of the median or quartiles; the feature classification is to classify features falling into different quadrants, namely: first quadrant: core key factors; second quadrant: controversial factors; third quadrant: stable irrelevant factors; fourth quadrant: noise factors.

[0035] Example threshold is 0.5 (adjustable): Quadrant 1: Core and Key Factors; Characteristics: High importance, high stability (e.g., MIS > 0.5, FCI > 0.5); Interpretation: All methods agree that it is of paramount importance and is the core driving factor of POSD; Decision: Prioritize retention for use in clinical prediction models or intervention guidelines; Second Quadrant: Controversial Factors; Characteristics: High importance, low stability (e.g., MIS > 0.5, FCI < 0.5); Interpretation: While the feature has high overall importance, there is significant disagreement among different methods in evaluating it. This usually indicates that the feature has complex interaction effects or only demonstrates its importance under specific model assumptions. Decision: In-depth diagnostic analysis should be conducted to explore the source of the discrepancy. These characteristics may reveal new biological mechanisms, but caution should be exercised when applying them directly to general models. Third quadrant: Stable irrelevant factors; Characteristics: low importance, high stability (e.g., MIS < 0.5, FCI > 0.5); Interpretation: All methods agree that it is unimportant; Decision: Remove from the model to reduce complexity and noise; Fourth quadrant: Noise factor; Characteristics: Low importance, low stability (e.g., MIS < 0.5, FCI < 0.5); Interpretation: Low importance and inconsistent assessments indicate this is typical noise. Decision: Prioritize elimination; The aforementioned method for identifying key factors of postoperative sleep disorders improves the robustness and interpretability of feature assessment by expanding method diversity, training details, and threshold adaptation, making it suitable for clinical decision support systems.

[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying key factors of postoperative sleep disorders, characterized in that, Includes the following steps: Step 1: Data preparation and preprocessing, including data acquisition and data preprocessing. Data acquisition involves obtaining multi-dimensional patient data from electronic medical record systems, monitoring devices, or clinical databases to form a candidate feature set. , where n is the total number of features, and whether or not sleep disorder occurs after surgery is determined as a binary classification label; data preprocessing cleans and transforms the raw data, including missing value imputation, outlier handling, data type conversion, and classification feature encoding; Step 2: Multi-source feature importance generation, including method selection, model training details, and importance calculation. Method selection involves pre-selecting at least five feature importance evaluation methods with different principles. Model training details include data splitting, hyperparameter tuning, handling class imbalance, and model evaluation. Importance calculation uses the selected methods to analyze the preprocessed data and calculate importance for each feature. Calculate importance, and output the result as a score or ranking; Step 3: Feature importance and stability fusion assessment, providing score-based and ranking-based analysis paths. The score-based analysis path includes score normalization and fusion index calculation, calculating the average importance score and feature consistency index; the ranking-based analysis path includes ranking acquisition and fusion index calculation, calculating the average ranking and feature importance stability index. Step 4: Four-quadrant analysis and key factor classification, including two-dimensional visualization and quadrant division and feature classification. The two-dimensional visualization constructs a scatter plot with importance index as the X-axis and stability index as the Y-axis. The quadrant division divides the plot into four quadrants according to the threshold and classifies the features.

2. The method for identifying key factors of postoperative sleep disorders according to claim 1, characterized in that, In the data collection of step one, the multidimensional data includes demographic characteristics, preoperative comorbidity characteristics, surgery and anesthesia-related characteristics, and postoperative management characteristics; the binary labels are determined by the Pittsburgh Sleep Quality Index or polysomnography.

3. The method for identifying key factors of postoperative sleep disorders according to claim 1, characterized in that, In the data preprocessing of step one, the missing value imputation uses the mean, median, or model prediction method; the classification feature encoding uses one-hot encoding or target encoding.

4. The method for identifying key factors of postoperative sleep disorders according to claim 1, characterized in that, In the method selection in step two, the five methods include global importance based on SHAP, importance based on permutation, built-in gain importance based on gradient boosting decision tree model, local importance based on LIME, and coefficient importance based on Lasso regression.

5. The method for identifying key factors of postoperative sleep disorders according to claim 1, characterized in that, In the model training details of step two, the data segmentation uses k-fold cross-validation; the hyperparameter tuning sets the learning rate, maximum depth, and number of iterations for the specific model; the handling of class imbalance uses oversampling or weight adjustment; and the model evaluation uses accuracy, precision, and recall.

6. The method for identifying key factors of postoperative sleep disorders according to claim 1, characterized in that, In the score-based analysis path of step three, the score is normalized as follows: for each method The resulting importance score vector Perform min-max normalization on all scores. Mapping to the [0,1] interval yields the normalized score. The formula is as follows: ， In the formula, The normalized score represents the characteristic. In the method The standardized importance value is given below, with a range of [0,1], where 0 represents the lowest importance and 1 represents the highest importance; The original importance scores represent the method. Features The initial evaluation score is directly output by the importance evaluation algorithm, reflecting the relative contribution of the feature in the model; For method The generated importance score vector of all features represents the method. For the entire feature set The set of fractions; The fusion index is calculated as follows: for each feature Based on its m normalized scores Calculate the following two core metrics: (1) Average Importance Score: This indicator measures the average importance level of a feature, and the formula is as follows: ， Its range is [0,1], and the larger the value, the higher the overall importance of the feature under multiple methods; (2) Feature Consistency Index: This index measures the stability and consistency of the conclusions regarding feature importance. The formula is as follows: ， In the formula, Features of The sample variance of the normalized scores. The value is 0.25, which is the theoretical maximum variance of the variable in the interval [0,1], and is used to ensure that the value range of FCI falls within the interval [0,1]. The closer the FCI value is to 1, the more consistent the different methods are in assessing the importance of this feature, and the more stable the conclusion; the closer the FCI value is to 0, the greater the divergence, and the less stable the conclusion.

7. The method for identifying key factors of postoperative sleep disorders according to claim 1, characterized in that, In the ranking-based analysis path of step three, the ranking is obtained as follows: for each method This yields a feature ranking list. , where the i-th feature The ranking is recorded as ; The fusion index is calculated as follows: for each feature Based on its Ranking Calculate the following two indicators: (1) Average ranking, calculated using the following formula: ， The smaller the average ranking value, the higher the overall importance. (2) Feature importance stability index, calculated using the following formula: ， In the formula, As a feature of The sample variance of each ranking. ,yes The theoretical maximum variance of the feature ranking [0,1] is used for normalization; The closer the FISI value is to 1, the more stable the ranking.

8. The method for identifying key factors of postoperative sleep disorders according to claim 1, characterized in that, In step four, quadrant division and feature classification, the threshold is defined by experts in the field or adaptively determined based on data distribution, including the use of median or quartiles; the feature classification is to classify features falling into different quadrants, namely: first quadrant: core key factors; second quadrant: controversial factors; third quadrant: stable irrelevant factors; fourth quadrant: noise factors.

Images