Method for predicting occurrence risk of postoperative shoulder pain
By integrating multiple risk factors and using genetic algorithms to screen features, a predictive factor library was constructed and transformed into a visualization tool, solving the problems of accuracy and universality in predicting postoperative pain risk in gynecological laparoscopic surgery, and achieving high-precision individualized assessment.
Patent Information
- Application Number
- CN202511730815.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-24
- Publication Date
- 2026-03-24
AI Technical Summary
In existing technologies, postoperative pain risk prediction models for gynecological laparoscopic surgery suffer from small sample sizes, low accuracy and generalizability, insufficient model validation, high costs, or reliance on a single variable, making it difficult to reflect population heterogeneity.
Integrating three major categories of risk factors—patient-related, surgical, and analgesic—this approach employs a genetic algorithm for intelligent feature selection, constructs a comprehensive predictive factor library, designs an adaptive genetic algorithm with dynamically adjusted weights, and utilizes a dataset binary search and iterative validation strategy to transform the data into an intuitive nomogram visualization tool that supports individualized risk assessment.
It improves the accuracy and clinical applicability of postoperative shoulder pain prediction, ensures the robustness and application value of the model in the real world, and provides a fast and accurate individualized risk assessment tool.
Smart Images
Figure CN121726044A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of postoperative shoulder pain risk prediction, and in particular to a method for predicting the risk of postoperative shoulder pain. BACKGROUND
[0002] Postoperative pain is a problem that patients are likely to encounter after surgery, which can cause great trouble to the patient's recuperation and subsequent treatment. Therefore, targeted prediction of postoperative pain risk means that the expected can be managed in advance, the patient can be mentally prepared, anxiety due to the unknown can be reduced, and the treatment plan before surgery can be assisted to choose the surgical plan with the lowest risk, which is convenient for the patient's subsequent rehabilitation.
[0003] Research on the prediction model of postoperative pain (Post-Iaparoscopic Shoulder Pain, PLSP) after gynecological laparoscopy mainly focuses on the field of postoperative chronic pain. Some researchers have constructed a prediction model of non-acute pelvic pain after gynecological laparoscopy, and have focused on the influence of psychological factors and pain sensitivity. The results support the significant role of psychological dimensions in the occurrence of chronic pain. However, the small sample size of this study may limit the universality of the conclusions and make it difficult to fully reflect the heterogeneity of the population. In addition, the short-form McGill pain questionnaire (SF-MPQ) used in the study takes a long time to evaluate and is limited in the application of acute pain situations. Secondly, some researchers have developed a postoperative pain prediction model based on SCN9A gene polymorphism. Through genotyping technology, 10 SNPs (p<0.05) were identified as being significantly related to postoperative pain intensity, moderate-to-severe pain occurrence and the need for self-controlled analgesia. The constructed tool shows good prediction performance, but this model only relies on genetic markers and does not integrate clinical and sociodemographic variables. In addition, genetic testing is costly and time-consuming, which limits its value in routine perioperative evaluation. In addition, some researchers have constructed a chronic pain risk prediction model for patients with endometrial cancer after robot-assisted laparoscopic hysterectomy. They found that preoperative pelvic pain is an independent risk factor for postoperative chronic pain. This study controlled the cancer stage to reduce the confounding bias caused by cancer pain, but since it was developed based on a single institution cohort and not externally validated, the universal applicability and accuracy of the model lack verification.
[0004] That is, the current prediction of postoperative pain risk after gynecological laparoscopy at least has the following problems: 1) There are too few special models, and the accuracy and universality are low, making it difficult to fully reflect the heterogeneity of the population; 2) There is a lack of universal validation when using other related models, and other related models have limitations such as single variable type, insufficient sample representation or insufficient validation. SUMMARY
[0005] To solve the above problems, the purpose of the present application is to provide a postoperative shoulder pain risk prediction method, which integrates patient self, surgery and analgesia three risk factors, builds a comprehensive prediction factor library, intelligently screens features through genetic algorithm with AUC maximization as the target, effectively removes redundancy and dimensionality, ensures the scientificity and optimality of the core variable combination, lays a solid foundation for the model; design a dynamic weight adjustment adaptive genetic algorithm to balance the performance and complexity of the model, use dataset dichotomy and cycle verification strategy to strictly evaluate the discrimination and calibration of the model, ensure that the final nomogram has high prediction accuracy and good clinical practicability; the complex model is converted into an intuitive nomogram visualization tool to support clinicians to quickly perform individualized risk quantitative evaluation, the preset performance threshold and cycle optimization mechanism form a closed loop of development-verification-optimization, which guarantees the robustness and application value of the model in the real world.
[0006] The technical scheme is realized by the following technical scheme: A postoperative shoulder pain risk prediction method, comprising the following steps: S1, obtaining the risk factors of each patient sample after shoulder pain, the risk factors including patient self sub-factors, surgery sub-factors and analgesia sub-factors, and simultaneously counting the actual shoulder pain incidence rate of each patient sample; S2, dividing each risk factor in step S1 into dataset one and dataset two, using GA algorithm to process dataset one with AUC maximization as the target to search for the optimal variable combination; S3, performing single factor analysis and multi-factor Logistic regression on each risk factor screened out by the optimal variable combination to obtain each core sub-factor; S4, outputting the corresponding nomogram model according to each core sub-factor and each actual shoulder pain incidence rate; the nomogram model at least includes multiple core sub-factors, Points axis and risk axis, the Points axis is a score axis, the risk axis is a shoulder pain rate axis corresponding to the actual shoulder pain incidence rate, each core sub-factor corresponds to a coordinate axis respectively under the condition of referring to the same Points axis; a TotalPoints axis is constructed based on the sum of scores of each core sub-factor on the Points axis; the sum of each score corresponding to each patient sample is predicted on the risk axis to obtain the corresponding shoulder pain incidence rate; S5, inputting dataset two into the nomogram model to calculate the predicted shoulder pain incidence rate of each patient sample, and calculating the AUC2 and calibration curve of the nomogram model in dataset two according to the predicted shoulder pain incidence rate and the actual shoulder pain or not in the dataset; When the AUC2 is not lower than the preset threshold and the calibration curve meets the calibration slope range, the nomogram model is used for postoperative shoulder pain prediction of new patients; when the AUC2 is lower than the preset threshold or the calibration curve does not meet the calibration slope range, the step of dividing the data set one and the data set two in step S2 is returned to cycle until the maximum cycle number is reached or the nomogram model is used for postoperative shoulder pain prediction of new patients.
[0007] Preferably, the patient self-sub-factors at least include age, BMI, whether severe anxiety exists before surgery, whether a history of pain outside the shoulder exists before surgery, whether pelvic adhesion exists before surgery, the surgical sub-factors at least include operation duration, operation type, bleeding volume, whether a peritoneal drainage tube is placed, and the analgesic sub-factors include whether an intravenous analgesic pump is used after surgery. Through multi-dimensional factor analysis, the complex causes of postoperative shoulder pain can be more accurately captured, and the prediction accuracy and clinical applicability of the model are improved.
[0008] Preferably, the judgment method of whether severe anxiety exists before surgery is as follows: the anxiety subscale in the Amsterdam Preoperative Anxiety and Information Scale is used to evaluate the preoperative anxiety level; wherein the anxiety subscale uses the scoring method, and the score reaching the corresponding score value is regarded as existing severe anxiety. The clear score threshold ensures the objectivity of the evaluation result, provides a reliable psychological factor quantitative index for the model, and enhances the scientificity of the prediction.
[0009] Preferably, the core sub-factors at least include age, BMI, whether severe anxiety exists before surgery, operation duration, and whether an intravenous analgesic pump is used after surgery. From numerous risk factors, five core factors with the highest prediction value are selected, achieving a balance between model simplification and performance optimization. These factors are convenient to obtain clinically, are convenient for practical application, and at the same time ensure that the prediction performance of the model is not affected.
[0010] Preferably, each core sub-factor is respectively provided with different score upper limits in each coordinate axis corresponding to the core sub-factor, and each score upper limit is in turn as follows from large to small: operation duration, BMI, age, whether an intravenous analgesic pump is used after surgery, and whether severe anxiety exists before surgery. According to the differences in clinical importance of various factors, different score upper limits are set, which reflects the relative weight relationship between factors. Operation duration obtains the highest score weight, which is consistent with the clinical cognition that it is an important surgical stress indicator, so that the nomogram score is more consistent with the clinical reality.
[0011] Preferably, when using the GA algorithm, the GA algorithm is set with an adaptive degree function F(x) = ω1×a(t)×E1(x)-ω2×b(t)×C(x)-ω3×c(t)×E2(x), where a(t), b(t), and c(t) are weight adjustment coefficients that dynamically change with the number of generations t, E1 corresponds to the AUC value, C corresponds to the number of sub-factors, E2 corresponds to the calibration error term, and ω1, ω2, and ω3 are the preset inertia weights of E1, C, and E2, respectively. When obtaining each initial population individual of the genetic algorithm, the number of initial population individuals is first adjusted using the adaptive degree function; then, the probability of selection P(xi) is set for each adjusted initial population individual, P(xi) = [F(xi) / [F(xj)]×[1+δ×|F(xi)-F_population average fitness| / F_population average fitness], where n corresponds to the initial population size, δ is the adjustment factor, and i and j are both positive integers. Multiple individuals are selected from each initial population individual after adaptive adjustment using the individual selection probability P(xi) to participate in crossover and mutation operations. Various crossover and mutation strategies are used to mutate each selected individual, obtaining each variable combination. The adaptive function comprehensively balances model performance, complexity, and error, achieving multi-objective optimization. The dynamic weight adjustment mechanism and fitness-based selection probability formula improve the algorithm's convergence efficiency and variable combination quality, ensuring the acquisition of the optimal feature subset.
[0012] Preferably, during the mutation operation, the mutated gene value is denoted as x'j. , where N(0, σ 2 () represents a mean of 0 and a variance of σ. 2 Normally distributed random numbers, where θ is a scaling factor, and UB represents X. j The corresponding upper bound value, LB, represents X. j The corresponding lower bound is r1, where r1 is a selected random number, and Δ represents a dynamically variable asynchronous long function. A normal distribution mutation operator is introduced, and the mutation amplitude is finely controlled through a scaling factor, enhancing the algorithm's global search capability. This intelligent mutation strategy effectively avoids local optima, promotes population diversity, and improves the efficiency of exploring variable combinations.
[0013] Preferably, when performing univariate analysis in step S3, the P-value of each risk factor is calculated separately.
[0014] Preferably, in step S3, when performing multivariate logistic regression, a specific threshold for the p-value is set, and each risk factor that meets the specific threshold from the univariate analysis results is selected for binary logistic regression. This two-step analysis strategy, first univariate and then multivariate, ensures that all variables included in the multivariate analysis are statistically significant.
[0015] The beneficial effects of this invention compared to the prior art are: The technical solution of this invention integrates three major categories of risk factors: patient-related, surgical, and analgesic factors, constructing a comprehensive predictive factor library. It employs a genetic algorithm to intelligently screen features with the goal of maximizing AUC, effectively removing redundancy and reducing dimensionality, ensuring the scientific rigor and optimality of the core variable combinations, thus laying a solid foundation for the model. An adaptive genetic algorithm with dynamically adjusted weights is designed to balance model performance and complexity. A binary dataset and iterative validation strategy are used to rigorously evaluate the model's discriminative and calibrated properties, ensuring that the final nomogram possesses both high predictive accuracy and good clinical applicability. The complex model is transformed into an intuitive nomogram visualization tool, supporting clinicians in quickly conducting individualized quantitative risk assessments. Preset performance thresholds and iterative optimization mechanisms form a closed loop of development-validation-optimization, guaranteeing the model's robustness and application value in the real world. Attached Figure Description
[0016] Figure 1 A flowchart for a method to predict the risk of postoperative shoulder pain; Figure 2 This is a diagram illustrating a nodal chart model. Detailed Implementation
[0017] The following will be based on embodiments of the present invention. Figure 1 and Figure 2 The technical solutions in the embodiments of the present invention will be described in detail below.
[0018] like Figure 1 The diagram shows a flowchart of a method for predicting the risk of postoperative shoulder pain. By integrating risk factors and using a genetic algorithm for intelligent feature selection, a prediction model with both high discriminative power and calibration is constructed. A rigorous dataset validation and iterative optimization mechanism is adopted to ensure the model's stability and reliability. Finally, the risk is visualized through an intuitive nomogram, which also provides an effective reference for clinical prediction of whether patients will experience postoperative shoulder pain.
[0019] The method specifically includes the following steps: S1. Obtain the risk factors for postoperative shoulder pain for each patient sample. The risk factors include patient-specific sub-factors, surgical sub-factors, and analgesia sub-factors. At the same time, calculate the actual incidence of shoulder pain for each patient sample.
[0020] Patient-specific sub-factors include at least age, BMI, preoperative history of severe anxiety, history of pain outside the shoulder, and preoperative pelvic adhesions. Surgical sub-factors include at least operative duration, surgical type, blood loss, and whether an abdominal drainage tube was placed. Analgesia sub-factors include whether an intravenous analgesia pump was used postoperatively. Multidimensional factor analysis can more accurately capture the complex causes of postoperative shoulder pain, improving the model's predictive accuracy and clinical applicability.
[0021] The method for determining the presence of severe preoperative anxiety is as follows: The anxiety subscale of the Amsterdam Preoperative Anxiety and Information Scale is used to assess preoperative anxiety levels. The anxiety subscale employs a rating scale, with scores reaching the corresponding thresholds considered as indicating severe anxiety. Clearly defined rating thresholds ensure the objectivity of the assessment results, provide reliable quantitative indicators of psychological factors for the model, and enhance the scientific rigor of the predictions.
[0022] S2. For each risk factor in step S1, divide it into dataset 1 and dataset 2. Use the GA algorithm to process dataset 1 with the goal of maximizing AUC and search for the optimal combination of variables.
[0023] When using the GA algorithm, the GA algorithm sets an adaptive function F(x) = ω1×a(t)×E1(x)-ω2×b(t)×C(x)-ω3×c(t)×E2(x), where a(t), b(t), and c(t) are weight adjustment coefficients that dynamically change with the number of generations t, E1 corresponds to the AUC value, C corresponds to the number of sub-factors, E2 corresponds to the calibration error term, and ω1, ω2, and ω3 are the preset inertia weights of E1, C, and E2, respectively. When obtaining each initial population individual of the genetic algorithm, the number of initial population individuals is first adjusted using the adaptive function; then, the probability of selection P(xi) is set for each adjusted initial population individual, P(xi) = [F(xi) / F(xj)]×[1+δ×|F(xi)-F_population average fitness| / F_population average fitness], where n corresponds to the initial population size, j=1, 2, 3...n, δ is the adjustment factor, and i and j are both positive integers. Multiple individuals are selected from each initial population individual after adaptive adjustment using the individual selection probability P(xi) to participate in crossover and mutation operations. Various crossover and mutation strategies are used to mutate each selected individual, obtaining each variable combination. The adaptive function comprehensively balances model performance, complexity, and error, achieving multi-objective optimization. The dynamic weight adjustment mechanism and fitness-based selection probability formula improve the algorithm's convergence efficiency and variable combination quality, ensuring the acquisition of the optimal feature subset.
[0024] During mutation operations, the mutated gene value is denoted as x'j. , where N(0, σ 2 () represents a mean of 0 and a variance of σ. 2 Normally distributed random numbers, θ is the scaling factor, UB stands for Upper Bound, UB represents X j The corresponding upper bound, LB stands for Lower Bound, where LB represents X. j The corresponding lower bound is defined by r1, which is a selected random number. Δ represents a dynamically variable asynchronous long function, if indicates "if," and otherwise indicates "otherwise." A normal distribution mutation operator is introduced, and the mutation amplitude is finely controlled through a scaling factor, enhancing the algorithm's global search capability. This intelligent mutation strategy effectively avoids local optima, promotes population diversity, and improves the efficiency of exploring variable combinations.
[0025] S3. For each risk factor selected from the optimal variable combination, perform univariate analysis and multivariate logistic regression to obtain each core sub-factor. Core sub-factors include at least age, BMI, presence of preoperative severe anxiety, operation duration, and whether an intravenous analgesia pump was used postoperatively. From numerous risk factors, five core factors with the highest predictive value were selected, achieving a balance between model simplification and performance optimization. These factors are clinically readily available and easy to apply, while ensuring that the model's predictive efficacy is not affected.
[0026] When performing univariate analysis, the p-value for each risk factor is calculated individually. When performing multivariate logistic regression, a specific threshold for the p-value is set, and each risk factor that meets the specific threshold from the univariate analysis results is selected for binary logistic regression. This two-step analysis strategy, first univariate and then multivariate, ensures that all variables included in the multivariate analysis are statistically significant. For example, the univariate analysis is as follows: Table 1: Univariate Analysis Table
[0027] Then, based on the results of the univariate analysis, the six variables with p-values less than 0.05 were included in the binary logistic regression analysis, and the results are as follows: Table 2: Results of Binary Logistic Regression Analysis
[0028] It should be noted that Beta represents the coefficient, SE represents the standard error, and Wald X represents the standard error. 2 This represents the Waldka square value, OR is the odds ratio, and 95% CI represents the 95% confidence interval.
[0029] S4. Based on each core sub-factor and the corresponding actual shoulder pain incidence rate, output the corresponding nomogram model.
[0030] likeFigure 2 The diagram shown is a nomogram model. A nomogram model includes at least multiple core sub-factors, a Points axis, and a risk axis. The Points axis represents the score axis, and the risk axis represents the shoulder pain rate corresponding to the actual shoulder pain incidence rate. Each core sub-factor corresponds to a separate coordinate axis when using the same Points axis as a reference. The Total Points axis is constructed based on the sum of the scores of each core sub-factor on the Points axis. For each patient sample, the sum of each score is used to predict the corresponding shoulder pain incidence rate on the risk axis.
[0031] Each core sub-factor has a different upper limit for its score on each corresponding coordinate axis. These upper limits, ranked from highest to lowest, are: surgical duration, BMI, age, whether an intravenous analgesia pump was used post-operatively, and whether severe preoperative anxiety was present. Setting different upper limits based on the clinical importance of each factor reflects the relative weighting of these factors. Surgical duration receives the highest weighting, consistent with its clinical recognition as an important indicator of surgical stress, making the nomogram score more aligned with clinical reality.
[0032] S5. Substitute Dataset 2 into the nomogram model, calculate the predicted incidence of each shoulder pain for each patient sample, and calculate the AUC2 and calibration curve of the nomogram model in Dataset 2 based on the predicted incidence of each shoulder pain in the dataset and whether each shoulder pain actually occurred.
[0033] When AUC2 is not lower than the preset threshold and the calibration curve meets the calibration slope range, the nomogram model is used for postoperative shoulder pain prediction of new patients; when AUC2 is lower than the preset threshold or the calibration curve does not meet the calibration slope range, the process returns to step S2 to divide dataset one and dataset two and repeats until the maximum number of iterations or the nomogram model is used for postoperative shoulder pain prediction of new patients.
[0034] In summary, this invention integrates three major categories of risk factors: patient-related, surgical, and analgesic factors, constructing a comprehensive predictive factor library. It employs a genetic algorithm to intelligently select features with the goal of maximizing AUC, effectively removing redundancy and reducing dimensionality, ensuring the scientific rigor and optimality of the core variable combinations, thus laying a solid foundation for the model. An adaptive genetic algorithm with dynamically adjusted weights is designed to balance model performance and complexity. A binary dataset and iterative validation strategy are used to rigorously evaluate the model's discriminative and calibrative properties, ensuring that the final nomogram possesses both high predictive accuracy and good clinical applicability. The complex model is transformed into an intuitive nomogram visualization tool, supporting clinicians in quickly conducting individualized quantitative risk assessments. Preset performance thresholds and iterative optimization mechanisms form a closed loop of development-validation-optimization, guaranteeing the model's robustness and application value in the real world, demonstrating significant advancements.
[0035] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.
Claims
1. A method for predicting the risk of postoperative shoulder pain, characterized in that, Includes the following steps: S1. Obtain the risk factors for postoperative shoulder pain for each patient sample. The risk factors include patient-specific sub-factors, surgical sub-factors, and analgesia sub-factors. At the same time, calculate the actual incidence of shoulder pain for each patient sample. S2. For each risk factor in step S1, divide it into dataset 1 and dataset 2. Use the GA algorithm to process dataset 1 with the goal of maximizing AUC and search for the optimal combination of variables. S3. Perform univariate analysis and multivariate logistic regression on each risk factor selected by the optimal variable combination to obtain each core sub-factor; S4. Based on each core sub-factor and its corresponding actual shoulder pain incidence rate, output the corresponding nomogram model. The nomogram model includes at least multiple core sub-factors, a Points axis, and a risk axis. The Points axis is the score axis, and the risk axis is the shoulder pain rate axis corresponding to the actual shoulder pain incidence rate. Each core sub-factor corresponds to a separate coordinate axis when the same Points axis is used as a reference. Construct the Total Points axis based on the sum of the scores of each core sub-factor on the Points axis. For each patient sample, predict the corresponding shoulder pain incidence rate on the risk axis based on the sum of the scores of each sub-factor. S5. Substitute Dataset 2 into the nomogram model, calculate the predicted incidence of each shoulder pain for each patient sample, and calculate the AUC2 and calibration curve of the nomogram model in Dataset 2 based on the predicted incidence of each shoulder pain in the dataset and whether each shoulder pain actually occurred. When AUC2 is not lower than the preset threshold and the calibration curve meets the calibration slope range, the nomogram model is used for postoperative shoulder pain prediction of new patients; when AUC2 is lower than the preset threshold or the calibration curve does not meet the calibration slope range, the process returns to step S2 to divide dataset one and dataset two and repeats until the maximum number of iterations or the nomogram model is used for postoperative shoulder pain prediction of new patients.
2. The method for predicting the risk of postoperative shoulder pain according to claim 1, characterized in that, Patient-specific factors include at least age, BMI, preoperative history of severe anxiety, preoperative history of pain outside the shoulder, and preoperative pelvic adhesions. Surgical factors include at least the duration of surgery, type of surgery, amount of blood loss, and whether an abdominal drainage tube is placed. Analgesia factors include whether an intravenous analgesia pump is used after surgery.
3. The method for predicting the risk of postoperative shoulder pain according to claim 2, characterized in that, The method for determining the presence of severe anxiety before surgery is as follows: The anxiety subscale of the Amsterdam Preoperative Anxiety and Information Scale is used to assess the level of preoperative anxiety; the anxiety subscale is scored, and a score reaching the corresponding value is considered to indicate the presence of severe anxiety.
4. The method for predicting the risk of postoperative shoulder pain according to claim 2, characterized in that, Core sub-factors include at least age, BMI, presence of severe anxiety before surgery, duration of surgery, and use of an intravenous analgesia pump after surgery.
5. The method for predicting the risk of postoperative shoulder pain according to claim 4, characterized in that, Each core sub-factor has a different upper limit for its score on each corresponding coordinate axis. The upper limits for each score, in descending order, are: operation time, BMI, age, whether an intravenous analgesia pump was used after the operation, and whether there was severe anxiety before the operation.
6. The method for predicting the risk of postoperative shoulder pain according to claim 1, characterized in that, When using the GA algorithm, the GA algorithm sets an adaptive function F(x) = ω1×a(t)×E1(x)-ω2×b(t)×C(x)-ω3×c(t)×E2(x), where a(t), b(t), and c(t) are all weight adjustment coefficients that change dynamically with the number of generations t, E1 corresponds to the AUC value, C corresponds to the number of sub-factors, E2 corresponds to the calibration error term, and ω1, ω2, and ω3 are the preset inertia weights of E1, C, and E2, respectively. When obtaining each initial population individual for the genetic algorithm, the number of initial population individuals is first adjusted using an adaptive function; then, the selection probability P(xi) is set for each adjusted initial population individual, where P(xi) = [F(xi) / F(xj)]×[1+δ×|F(xi)-F average fitness| / F average fitness], where n corresponds to the initial population size, δ is the adjustment factor, and i and j are both positive integers. Multiple individuals are selected from each initial population individual after fitness adjustment using the individual selection probability P(xi) to participate in the crossover and mutation operation. Various crossover and mutation strategies are used to mutate each selected population individual to obtain each variable combination.
7. The method for predicting the risk of postoperative shoulder pain according to claim 6, characterized in that, During mutation operations, the mutated gene value is denoted as x'j. , where N(0, σ 2 () represents a mean of 0 and a variance of σ. 2 Normally distributed random numbers, where θ is a scaling factor, and UB represents X. j The corresponding upper bound value, LB, represents X. j The corresponding lower bound value, r1 is the selected random number, and Δ represents a dynamically variable asynchronous long function.
8. The method for predicting the risk of postoperative shoulder pain according to claim 1, characterized in that, In step S3, when performing univariate analysis, the p-value for each risk factor is calculated separately.
9. The method for predicting the risk of postoperative shoulder pain according to claim 1, characterized in that, When performing multivariate logistic regression in step S3, a specific threshold for the P-value is set, and each risk factor that meets the specific threshold in the results of the univariate analysis is selected for binary logistic regression.