Suicide risk prediction method and system for juvenile depression

Abstract

The invention relates to the field of medical artificial intelligence, and discloses a teenager depression suicide risk prediction method and system, and the method comprises the steps: selecting teenagers which are 13-20 years old, have severe depression, and do not take drugs for at least 3 months; suicide ideas and behaviors are evaluated by adopting a Columbia suicide severity evaluation scale, participants are divided into a suicide-free group and a non-suicide-free group, big five personality characteristics are evaluated by applying a big five personality scale, and child-stage abuse conditions are evaluated by utilizing a child-stage wound questionnaire; the participants are randomly divided into a training queue and a verification queue according to the proportion of 7: 3; screening suicide prediction factors by adopting t test, rank sum test and chi-square test in combination with single-factor logistic regression, random forest and lasso regression models, and finally determining six optimal variables; constructing a suicide parasite risk prediction column line graph; and the prediction accuracy, the fitting degree, the clinical applicability and the generalization ability of the column graph model are evaluated. The method is accurate in prediction, systematic and efficient, and high in clinical practicability.

Description

Technical Field

[0001] The present invention relates to the fields of psychiatry, psychology, epidemiology and artificial intelligence, and in particular to a method and system for predicting the suicide risk of adolescent patients with depression. Background Art

[0002] In recent years, the incidence of adolescent depression has been on the rise, and suicide has become a major cause of death among adolescents. Accurately predicting the suicide risk of adolescents with depression is crucial for timely intervention and saving lives.

[0003] Although there have been relevant studies on the prediction of adolescent suicide risk, for example, patent CN104881719A discloses a method for constructing an adolescent suicide or self-injury risk assessment and warning model, and patent CN114255871A discloses a method for generating an adolescent suicide risk prediction model and a prediction system, both of which can predict adolescent suicide risk to a certain extent, but they cannot be directly linked to depression and can only be applied to certain specific adolescent groups, and cannot be generalized to all adolescent groups. More importantly, the prediction results often have large errors compared to the actual situation.

[0004] Currently, the prediction of suicide risk among adolescents with depression still faces many challenges.

[0005] On the one hand, existing prediction tools and methods lack accuracy and reliability. For example, some tools rely solely on a single dimension of assessment and fail to fully consider the multiple factors that influence suicide risk. Some tools also have limited applicability across different cultural backgrounds and populations, leading to significant deviations in prediction results.

[0006] On the other hand, existing studies are prone to overfitting problems during variable screening and model construction, which makes the model's generalization ability poor and difficult to be widely used in actual clinical scenarios.

[0007] Therefore, developing an efficient, accurate and widely applicable method and system for predicting suicide risk among adolescents with depression has important practical significance. Summary of the Invention

[0008] The present invention aims to provide a method for predicting suicide risk in adolescents with depression, so as to solve the problem of the lack of systematic tools for predicting suicide risk in adolescents with depression.

[0009] In order to solve the above problems, the present invention adopts the following technical solutions:

[0010] Solution 1: A method for predicting suicide risk in adolescents with depression, including the following steps:

[0011] Step 1: Identify the research subjects. Adolescents aged 13 to 20 years, diagnosed with major depression, not taking medication for at least 3 months, with a total score of ≥17 on the 17-item Hamilton Depression Rating Scale and ≥40 on the Children's Depression Rating Scale-Revised were selected. Individuals with severe physical illness, a history of epileptic seizures and organic brain diseases, schizophrenia, and psychoactive substance abuse were excluded.

[0012] Step 2: Using an interview questionnaire and the Columbia-Suicide Severity Rating Scale to assess suicidal thoughts and behaviors, participants were divided into a suicide attempt group and a non-suicide attempt group. The Big Five Personality Inventory was used to assess Big Five personality traits, and the Childhood Trauma Questionnaire was used to assess childhood abuse.

[0013] Step 3: Participants were randomly divided into a training cohort and a validation cohort in a 7:3 ratio. Descriptive analyses were performed using R software. Statistical analysis was performed using t-tests, rank-sum tests, and chi-square tests, combined with univariate logistic regression, random forest, and lasso regression models to identify six optimal predictors of suicide.

[0014] Step 4: Use the six optimal variables to construct a suicide attempt risk prediction nomogram. Assign a preliminary graphic score from 0 to 100 to each predictor. The scores are determined by the predictor line and summed to obtain the total score. The total score corresponds to the position of the "total score" axis, and the bottom scale reflects the probability of suicide attempt.

[0015] Step 5: The receiver operating characteristic curve, area under the ROC curve, calibration curve, Hosmer-Lemeshow test, and decision curve analysis were used to evaluate the predictive accuracy, goodness of fit, clinical applicability, and generalization ability of the nomogram model.

[0016] Beneficial effects: Clarify the research subject standards to ensure the homogeneity and pertinence of the research sample; collect data from multiple dimensions to comprehensively obtain the factors affecting suicide risk; combine multiple statistical methods to screen variables to improve the accuracy and reliability of predictive factors; construct a nomogram model to intuitively present the suicide risk probability; and use multiple evaluation methods to ensure good model performance and provide a scientific basis for suicide risk prediction.

[0017] The solution of the present invention focuses on adolescent patients with depression. By accurately collecting data in multiple dimensions, scientifically screening predictive factors to construct a nomogram model and comprehensively evaluate the model performance, it also constructs a fully functional prediction system with the advantages of accurate prediction, efficient system and strong clinical practicality. It can effectively assist in the assessment and intervention of suicide risks in adolescents with depression.

[0018] Preferably, the univariate logistic regression is performed on all samples of the training cohort to screen out variables that are significantly correlated with suicide attempts, and is visualized using a forest plot.

[0019] Beneficial effects: The forest plot visually displays the association between variables and suicide attempts, allowing researchers to quickly understand important factors, providing a clear reference for subsequent variable screening and improving screening efficiency and accuracy.

[0020] Preferably, the random forest analysis reflects the contribution of each independent variable to the level of suicide attempts by generating an average Gini coefficient reduction, and the top ten variables are selected in descending order of contribution value. During the analysis process, it is determined that the estimated error rate of the out-of-bag sample is the lowest when the number of decision trees is 800 and the number of variables is 4.

[0021] Beneficial effects: By leveraging the advantages of the random forest algorithm, the complex relationships between variables are effectively handled, the influence of interfering factors is reduced, and the selected parameters make the model more stable and accurate, ensuring that the selected variables are of great value in predicting suicide attempts.

[0022] Preferably, the lasso regression selects optimal parameters through ten-fold cross validation, screens predictive factors of suicide attempts from potential 22 factors, reduces the correlation between variables, and avoids model overfitting.

[0023] Beneficial effects: Ten-fold cross-validation improves the scientificity and reliability of parameter selection, reduces variable correlation and avoids overfitting, makes the model more generalizable, and can stably and accurately predict suicide risk in different data samples.

[0024] Preferably, the six optimal variables are Hamilton Depression Rating Scale score, Patient Health Questionnaire Somatic Symptom Cluster Scale score, Beck Suicidal Ideation Scale score, extraversion, emotional neglect, and physical neglect.

[0025] Beneficial effects: These six variables are selected from multiple key dimensions such as clinical symptoms, suicidal ideation, personality traits, and childhood trauma. They can comprehensively and accurately reflect the relevant factors of suicide risk in adolescent patients with depression, and provide core variables for building an efficient and accurate prediction model.

[0026] Solution 2: The present invention also provides a suicide risk prediction system for adolescent depression, comprising:

[0027] The data collection module is used to collect relevant data of adolescent patients with depression, including basic information, diagnostic information, and various scale assessment data;

[0028] The data analysis module uses a variety of statistical analysis methods to process and analyze the collected data and screen out key predictive factors;

[0029] Model construction module, which constructs a suicide risk prediction nomogram model based on the screened predictive factors and optimizes and adjusts the model;

[0030] The risk assessment module substitutes the patient's specific data into the nomogram model to calculate the probability of suicide attempt and assess the patient's suicide risk level.

[0031] Beneficial effects: The data acquisition module ensures comprehensive and accurate data, providing a basis for analysis; the data analysis module efficiently screens key factors; the model construction module builds and optimizes the model to improve prediction accuracy; the risk assessment module quickly and accurately assesses the risk level, providing a direct basis for clinical intervention.

[0032] Preferably, the data acquisition module is designed with a special data entry interface, which has mandatory item prompts and data format checking functions.

[0033] Beneficial effects: Improve the accuracy and completeness of data entry, reduce data errors and omissions, and ensure the reliability of subsequent analysis and prediction.

[0034] Preferably, the data analysis module integrates the analysis functions of SPSS and R software to realize an automated analysis process.

[0035] Beneficial effects: Integrate professional statistical software functions, improve data analysis efficiency and accuracy, reduce manual operation errors, quickly obtain analysis results, and provide strong support for model construction.

[0036] Preferably, the model building module uses professional drawing tools to build the nomogram model and provides model parameter adjustment and optimization functions.

[0037] Beneficial effects: The nomograms constructed with professional drawing tools are more accurate and intuitive, and the parameter adjustment and optimization functions enable the model to adapt to different data and research needs, improving the practicality and accuracy of the model.

[0038] Preferably, the risk assessment module displays risk levels in an intuitive manner, including low risk, medium risk, and high risk.

[0039] Beneficial effects: It helps medical staff to quickly understand and judge the patient's suicide risk level, take appropriate intervention measures in a timely manner, and improve clinical work efficiency and intervention effects.

[0040] Advantages compared to existing technologies:

[0041] High prediction accuracy: This invention comprehensively considers multiple factors of adolescent depression patients. Through a scientific and rigorous variable screening and model building process, the constructed nomogram model shows good discrimination and calibration in both training and validation cohorts, with a high area under the curve, and can more accurately predict suicide risk. Compared with the single-dimensional assessment or simple model construction in the existing technology, it greatly improves the accuracy of prediction.

[0042] The model has strong generalization capabilities: By combining univariate logistic regression, random forest, and lasso regression to screen variables, the model effectively avoids overfitting, giving it better generalization capabilities and enabling it to stably predict suicide risk across diverse sample data. However, some existing studies have not fully considered the overfitting issue, resulting in poor performance in practical applications.

[0043] Good clinical applicability: Decision curve analysis shows that the model of the present invention can generate net benefits within a wide range of probabilities, and the risk prediction system can intuitively display the risk level, making it easier for medical staff to understand and use it, providing a practical tool for clinical decision-making. Compared with some prediction methods in the existing technology that are difficult to apply in actual clinical scenarios, it has greater clinical value.

[0044] Comprehensive multi-dimensional assessment: Variables are screened from four aspects: demographic information, basic clinical variables, childhood trauma, and personality traits to comprehensively assess the risk of suicide attempts. Existing technologies may only focus on certain aspects and cannot fully reflect the factors affecting suicide risk.

[0045] Convenient and efficient operation: The suicide risk prediction system integrates data collection, analysis, model building and risk assessment functions. It is easy to operate and has a high degree of automation, which greatly shortens the assessment time and improves work efficiency. Existing technologies may require tedious manual operations and complex calculation processes, which are less efficient.

[0046] Precise research subjects and comprehensive data collection: This patent focuses on adolescents aged 13-20 diagnosed with major depression, unlike other comparable documents that target a broad range of adolescents or cancer patients. It also utilizes multiple professional scales, such as the Brief International Neuropsychiatric Interview for Children and Adolescents and the Revised Big Five Personality Inventory, to collect data on multiple dimensions, including mental state, personality traits, and childhood trauma. This provides richer, deeper, and more accurate information than can be obtained using custom questionnaires or simple surveys alone.

[0047] Scientific variable screening, rigorous model construction and evaluation: This patent combines univariate logistic regression, random forest, and lasso regression to screen variables, comprehensively considering multiple factors to determine key predictors. This is more scientific and comprehensive than using logistic regression alone or simple screening methods. A nomogram model is constructed and evaluated using multiple methods, including ROC curves, calibration curves, Hosmer-Lemeshow tests, and DCA, to comprehensively and rigorously verify model performance and ensure predictive accuracy, reliability, and clinical applicability. Existing technologies use relatively limited evaluation methods.

[0048] The system is fully functional and has high clinical application value: The suicide risk prediction system constructed in this patent covers data collection, analysis, model building, and risk assessment modules, each of which works closely together and has comprehensive functions. For example, the data collection module ensures data quality, while the risk assessment module intuitively displays risk levels and provides intervention recommendations. Its simple and efficient operation significantly improves clinical work efficiency, a feature not found in other comparable documents.

[0049] Breaking through the limitations of traditional research: Existing technologies for predicting adolescent suicide risk suffer from issues such as unclear research subjects, incomplete data collection, and limited model construction and evaluation methods. This patent precisely identifies the research subject, utilizes multi-dimensional data collection and multiple advanced analytical methods, and constructs a comprehensive prediction system. This is not a simple improvement or conventional application, but a significant breakthrough in traditional research approaches.

[0050] Solving a key problem in a specific field: Targeting the specific area of ​​suicide risk prediction for adolescents with depression, this patent effectively addresses the challenge of accurate predictions, a challenge that existing technologies struggle to address. By screening key variables to build a high-precision model, it provides a scientific basis for clinical intervention. This effective solution to a specific problem is not something that would be readily conceived by those skilled in the art based on existing technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a forest plot of the univariate logistic regression of the training cohort in Example 1 of the present invention, showing the degree of association and statistical significance between each variable in the univariate logistic regression analysis and suicide attempts, providing an intuitive basis for screening suicide prediction factors.

[0052] Figure 2 This is the variable screening process of the random forest in the training queue of embodiment 1 of the present invention. Figure 2 (A) Visualization of the relationship between the overall error rate and the number of trees, used to determine the appropriate number of trees in the random forest algorithm; Figure 2 (B) Assess the importance of all variables based on the Gini mean reduction and select key variables according to their importance.

[0053] Figure 3 This is the variable screening process of Lasso regression in the training queue of Example 1 of the present invention. Figure 3 (A) Lasso coefficient distribution diagram of 23 risk factors, showing the changes in the coefficients of each variable; Figure 3 (B) The optimal parameter (λ value) was selected through ten-fold cross-validation. The two vertical dashed lines are drawn based on the optimal scores under the minimum criterion and the 1 times standard error criterion, respectively, to screen out important variables.

[0054] Figure 4The common variables screened out by the Venn diagram in Example 1 of the present invention are shown. The intersection of the variables screened out by the three methods of univariate logistic regression, random forest and lasso regression is shown, and the six optimal variables finally used to construct the nomogram are determined.

[0055] Figure 5 This is a nomogram for predicting suicide attempts according to Example 1 of the present invention. It is constructed based on six optimal variables and the probability of a patient's suicide attempt can be calculated using the nomogram.

[0056] Figure 6 Receiver operating characteristic (ROC) curve of a nomogram according to an embodiment of the present invention. Figure 6 (A) is the training queue, Figure 6 (B) The validation cohort was used to evaluate the discrimination of the model using the ROC curve and the area under the curve (AUC).

[0057] Figure 7 This is the calibration curve of the nomogram in the study of Example 1 of the present invention. Figure 7 (A) is the training queue, Figure 7 (B) Validation cohort, used to observe the consistency between the predicted level of suicide attempts and the actual level observed.

[0058] Figure 8 This is a decision curve analysis (DCA) of a nomogram according to an embodiment of the present invention. Figure 8 (A) is the training queue, Figure 8 (B) For the validation cohort, DCA was used to evaluate the net benefit of the model at different threshold probabilities to determine the clinical applicability of the model.

[0059] Figure 9 This is a logic block diagram of embodiment 1 of the present invention. DETAILED DESCRIPTION

[0060] The following is further described in detail through specific implementation methods:

[0061] The reference numerals in the drawings of the specification include: control module 1 , data acquisition module 2 , data analysis module 3 , model building module 4 , and risk assessment module 5 .

[0062] Example 1

[0063] The suicide risk prediction method for adolescents with depression accurately predicts the suicide risk of adolescents with depression by collecting data from multiple dimensions, scientifically screening predictive factors, constructing a nomogram model, and comprehensively evaluating model performance. The specific steps are as follows:

[0064] Data Collection: Adolescents aged 13 to 20 years with a diagnosis of major depressive disorder according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DMS-5), who had not taken medication for at least three months and a total score of 17 or higher on the 17-item Hamilton Depression Rating Scale (HAMD-17) or 40 or higher on the Children's Depression Rating Scale-Revised (CDRS-R) were included in the study. Individuals with severe medical illness, a history of epileptic seizures or organic brain disease, psychiatric disorders such as schizophrenia, or alcohol or other psychoactive substance abuse were excluded. Assessments were performed using the Mini-International Neuropsychiatric Interview for Children and Adolescents (MINI-KID) and the Mini-International Neuropsychiatric Interview (MINI). The Columbia-Suicide Severity Rating Scale (C-SSRS) was used to assess suicidal thoughts and behaviors, and the participants were divided into a suicide attempt (SA) group and a non-suicide attempt (NSA) group. The Big Five Personality Inventory-Revised (NEO-FFI-R) was used to assess the Big Five personality traits. The Childhood Trauma Questionnaire (CTQ) was used to assess childhood abuse.

[0065] Feature Selection: All participants were randomly divided into a training cohort and a validation cohort at a 70% / 30% ratio. Descriptive analyses were performed using SPSS 26.0 software, and other statistical analyses were performed using R software (version 4.3.2). Student t-tests, rank-sum tests, and chi-square tests were used to ensure the feasibility of randomization. Univariate logistic regression, random forest, and lasso regression models were used to identify potential predictors of suicide.

[0066] Univariate logistic regression: Univariate logistic regression analysis was performed on all samples in the training cohort to screen out variables that were significantly associated with suicide attempts. The results are presented in Figure 1 In the forest plot of the univariate logistic regression of the training cohort, the degree of association and statistical significance between each variable and suicide attempt can be intuitively seen.

[0067] Random Forest: The random forest algorithm is used to generate the mean reduction in the Gini coefficient (MDG) to reflect the contribution of each independent variable to the level of suicide attempts. The appropriate number of decision trees (ntree) and number of variables (mtry) are determined through multiple tests and adjustments. Figure 2 During the variable screening process of random forest in the training cohort, Figure 2 (A) shows the visualization of the relationship between the overall error rate and the number of trees, which helps determine the appropriate number of trees; Figure 2 (B) The importance of all variables was assessed based on the Gini mean reduction, and the top ten variables were selected in descending order of MDG value.

[0068] Lasso regression: Use lasso regression to reduce the correlation between variables and avoid overfitting of the model. Figure 3During the variable screening process of Lasso regression in the training queue, Figure 3 (A) shows the distribution of the lasso coefficients of 23 risk factors, which clearly shows the changes in the coefficients of each variable; Figure 3 (B) The optimal parameter (λ value) was selected through ten-fold cross-validation. The two vertical dashed lines are drawn based on the optimal scores under the minimum criterion and the 1 times standard error criterion, respectively. Finally, predictors of suicide attempt (SA) were screened out from 22 potential factors.

[0069] Comprehensive screening: Combining the characteristic variables suggested by the above three methods, Figure 4 The common variables screened out by the Venn diagram eventually determined the six optimal variables, namely, Hamilton Depression Rating Scale (HAMD) score, Patient Health Questionnaire Somatic Symptom Cluster Scale (PHQ-15) score, Beck Suicidal Ideation Inventory (BSSI) score, extraversion, emotional neglect, and physical neglect.

[0070] Nomogram construction: The six optimal variables screened were used to construct a suicide attempt risk prediction nomogram, such as Figure 5 A nomogram for predicting suicide attempts is shown. Each predictor is assigned a preliminary graphical score ranging from 0 to 100. Scores are determined by the predictor line and summed to create a total score, which corresponds to the position on the "Total Score" axis. The bottom scale reflects the probability of suicide attempt. Healthcare professionals or researchers can find the corresponding score on the nomogram based on the patient's score on each scale and sum the scores to obtain an estimated probability of suicide attempt.

[0071] Model evaluation: The receiver operating characteristic (ROC) curve, area under the ROC curve (AUC), calibration curve, Hosmer-Lemeshow test, and decision curve analysis (DCA) were used to evaluate the predictive accuracy, goodness of fit, clinical applicability, and generalization ability of the nomogram model.

[0072] ROC curve and AUC: Draw the ROC curve of the training cohort and the validation cohort, the results are as follows Figure 6 The receiver operating characteristic (ROC) curve of the nomogram is shown as follows: Figure 6 (A) is the training queue, Figure 6 (B) shows the validation cohort. The larger the area under the ROC curve (AUC), the stronger the model's ability to discriminate between suicide attempts and non-suicide attempts. The AUC for the training cohort was 0.83 (95% confidence interval: 0.767-0.892), and the AUC for the validation cohort was 0.803 (95% confidence interval: 0.688-0.919), indicating that the model has good discrimination.

[0073] Calibration curve: The calibration curve is used to observe the agreement between the predicted level of suicide attempts and the actual level observed. Figure 7 is the calibration curve of the nomogram in this study, Figure 7 (A) is the training queue, Figure 7 (B) Validation cohort. The calibration curves in both cohorts showed bias correction, and the curves were significantly close to the ideal straight line, indicating that the predicted results were consistent with the actual results.

[0074] Hosmer-Lemeshow test: After the Hosmer-Lemeshow test, the P value of the training cohort was 0.519, and the P value of the validation cohort was 0.704. When the P value is greater than 0.05, it indicates that there is no difference in the probability distribution between the two groups, and the calibration result is ideal, that is, the model fit is good.

[0075] Decision Curve Analysis (DCA): Clinical applicability was assessed using decision curve analysis (DCA), which quantified the net benefit at different threshold probabilities. Figure 8 The decision curve analysis (DCA) of the nomogram showed the results of the training cohort and the validation cohort, where Figure 8 (A) is the training queue, Figure 8 (B) Validation cohort. The results showed that the model produced a net benefit with a wide range of probabilities, ranging from approximately 7% to 91%, in the training cohort and from 9% to 94% in the validation cohort, indicating that the model was beneficial in clinical decision-making.

[0076] Based on the above prediction method, the present invention constructs a set of Figure 9 The suicide risk prediction system shown includes a control module 1 and a data acquisition module 2, a data analysis module 3, a model building module 4 and a risk assessment module 5 respectively connected to the control module 1.

[0077] Data Collection Module 2: This module collects data related to adolescent depression, including basic information, diagnostic information, and various scale assessment data, providing comprehensive data support for subsequent analysis. This module features a dedicated data entry interface to facilitate accurate entry of patient data by medical staff, ensuring data integrity and accuracy.

[0078] Data Analysis Module 3: Utilizes a variety of statistical analysis methods to process and analyze the collected data and identify key predictive factors. This module integrates the analytical capabilities of SPSS and R software, automatically performing descriptive analysis, t-tests, rank-sum tests, chi-square tests, and complex analyses such as univariate logistic regression, random forest, and lasso regression on the entered data, quickly and accurately identifying important variables associated with suicide attempts.

[0079] Model Building Module 4: Based on the selected predictive factors, a suicide risk prediction nomogram model is constructed and continuously optimized and adjusted. This module uses professional drawing tools to construct the nomogram and provides model parameter adjustment and optimization capabilities. The nomogram can be refined to improve the model's predictive performance based on different datasets and research needs.

[0080] Risk Assessment Module 5: Plug the patient's specific data into the nomogram model to calculate the probability of a suicide attempt, intuitively assessing the patient's suicide risk level and providing a basis for clinical intervention. This module receives the results of Data Analysis Module 3 and, through simple operations, plugs the patient's data into the nomogram model to calculate the probability. It then intuitively displays the risk level, such as low risk, medium risk, and high risk, helping medical staff quickly determine the patient's suicide risk level and formulate appropriate intervention measures.

[0081] The specific implementation process is as follows:

[0082] Data Collection and Implementation Steps: The study was conducted in Chongqing. Through collaboration with local medical institutions and schools, adolescent patients with depression who met the inclusion criteria were widely recruited. After rigorous screening, 277 patients with major depressive disorder (MDD) aged 13 to 20 were selected as research subjects. These patients were diagnosed with major depressive disorder according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DMS-5), had not taken medication for at least 3 months, and met the criteria of a total score of ≥17 on the 17-item Hamilton Depression Rating Scale (HAMD-17) or a total score of ≥40 on the Children's Depression Rating Scale-Revised (CDRS-R).

[0083] All study subjects were comprehensively assessed by professionally trained investigators using the Mini-International Neuropsychiatric Interview for Children and Adolescents (MINI-KID) and the Mini-International Neuropsychiatric Interview (MINI). The Columbia-Suicide Severity Rating Scale (C-SSRS) was used to categorize participants into a suicide attempt (SA) group and a non-suicide attempt (NSA) group. The suicide attempt group included 79 patients (28.5% of the total sample) who had engaged in at least one suicidal behavior within the previous three months, while the non-suicide attempt group included 198 patients who had never attempted suicide in their lifetime.

[0084] Participants were instructed to complete the revised Big Five Personality Inventory (NEO-FFI-R) and the Childhood Trauma Questionnaire (CTQ). After completing the questionnaire collection, the data were initially collated and checked to ensure its completeness and accuracy, providing a reliable basis for subsequent analysis.

[0085] 2. Data Analysis and Model Construction Implementation Steps: SPSS 26.0 software was used to conduct a descriptive analysis of the collected data to obtain basic characteristics of the research subjects. For example, the average age of all samples was 16 years old, with 120 male patients (43.3%) and 157 female patients (56.7%).

[0086] Using R software (version 4.3.2), 70% of the participants (194 patients) were randomly assigned to the training cohort and 30% (83 patients) to the validation cohort. The characteristics of the training and validation cohorts were compared using t-tests, rank-sum tests, and chi-square tests. The results showed no statistically significant differences between the two groups in age, gender, and other characteristics (p>0.05), indicating that the randomization was feasible.

[0087] Feature selection was performed in the training cohort. First, predictors of suicide were screened by univariate logistic regression and forest plots were used ( Figure 1 ) for visual display. Figure 1 It can be seen that nine variables including self-harm, Hamilton Depression Rating Scale (HAMD) score, and Patient Health Questionnaire Somatic Symptom Cluster Scale (PHQ-15) score were significantly correlated with suicide attempts.

[0088] Then, we conduct random forest analysis. After multiple tests and adjustments, we find that when the number of decision trees (ntree) is 800 and the number of variables (mtry) is 4, the estimation error rate of out-of-bag samples is the lowest (out-of-bag error rate OBB = 25.77%). Figure 2 (B) Based on the Gini mean reduction, the importance of all variables was evaluated. It was found that ten variables, including the Beck Suicidal Ideation Inventory (BSSI) score and the Hamilton Depression Rating Scale (HAMD) score, were the most important.

[0089] Then, lasso regression analysis was performed, Figure 3 (A) Lasso coefficient distribution diagram of 23 risk factors and Figure 3 (B) The results of selecting the optimal parameter (λ value) through ten-fold cross-validation showed that 7 predictors of suicide attempt (SA) were finally screened out from the potential 22 factors.

[0090] Combining the three methods, through the Venn diagram ( Figure 4 ) screened out six optimal variables, namely Hamilton Depression Rating Scale (HAMD) score, Patient Health Questionnaire Somatic Symptom Cluster Scale (PHQ-15) score, Beck Suicide Ideation Inventory (BSSI) score, extraversion, emotional neglect, and physical neglect. Using these six optimal variables, a nomogram for predicting the risk of suicide attempts was constructed ( Figure 5 ).

[0091] 3. Model Evaluation Implementation Steps: The constructed nomogram model was comprehensively evaluated using the training and validation cohorts. The area under the curve (AUC) of the nomograms for the training and validation cohorts was calculated. The AUC for the training cohort was 0.83 (95% confidence interval: 0.767-0.892), and the AUC for the validation cohort was 0.803 (95% confidence interval: 0.688-0.919). This indicates that the model has good ability to distinguish between suicide attempts and non-suicide attempts. The closer the AUC is to 1, the better the discrimination effect. AUC values ​​of 0.83 and 0.803 indicate good discrimination.

[0092] In the training cohort, the specificity was 0.813, the sensitivity was 0.733, the Youden Index was 0.546, and the critical value was 0.366. In the validation cohort, the specificity was 0.875 and the sensitivity was 0.579. Specificity reflects the model's ability to correctly identify patients who did not attempt suicide. Both the training and validation cohorts had high specificity, indicating that the model's judgment of non-suicide attempt patients was relatively accurate. Sensitivity reflects the model's ability to correctly identify patients who did attempt suicide. Although the sensitivity of the validation cohort was relatively low, it was still within an acceptable range overall. The Youden Index combines specificity and sensitivity. The higher the value, the higher the diagnostic value of the model. A Youden Index of 0.546 indicates that the model has some diagnostic value.

[0093] Through the calibration curve ( Figure 7 ) observed consistency between the predicted and observed levels of suicide attempts. Calibration curves in both the training and validation cohorts demonstrated bias correction and were significantly close to ideal straight lines. This suggests that the model's predicted probability of suicide attempts closely matches the actual probability, indicating a high degree of reliability.

[0094] The Hosmer-Lemeshow test showed a P value of 0.519 for the training cohort and 0.704 for the validation cohort. A P value > 0.05 indicates no difference in the probability distribution between the two groups and an ideal calibration result, indicating a good model fit that accurately reflects the actual situation.

[0095] The results of decision curve analysis (DCA) showed that ( Figure 8 ). In the training cohort, the model produced a net benefit over a wide range of probabilities, ranging from approximately 7% to 91%. In the validation cohort, the net benefit was observed over a range of probabilities, ranging from 9% to 94%. This suggests that applying the model to assess suicide risk at different probability thresholds can provide valuable insights for physicians in clinical decision-making, helping them make more informed decisions and reducing unnecessary interventions and missed diagnoses.

[0096] Prediction System Implementation Steps: Develop a suicide risk prediction system based on the aforementioned prediction method. In Data Collection Module 2, design a concise and comprehensive data entry interface with features such as mandatory field prompts and data format checks to ensure that medical staff can accurately enter patient data, including basic information and scores from various scales.

[0097] Data Analysis Module 3 integrates the core analytical functions of SPSS and R software to automate the analysis process. Once data entry is complete, the system automatically calls the analysis program to quickly complete descriptive analysis, t-tests, rank sum tests, chi-square tests, and complex analyses such as univariate logistic regression, random forest, and lasso regression, effectively screening key predictive factors.

[0098] Model Building Module 4 utilizes specialized drawing tools to construct a precise nomogram model based on the six selected optimal variables. To meet diverse research needs and data variations, it also provides flexible model parameter adjustment and optimization capabilities, enabling researchers to tailor nomogram improvements and enhance model predictive performance.

[0099] Risk Assessment Module 5 receives the results from Data Analysis Module 3. Medical staff only need to input relevant patient data, and the system quickly inserts this data into a nomogram model to calculate the probability of suicide attempts. It then displays the risk levels in an intuitive and easy-to-understand manner, such as low risk (probability of suicide attempts <30%), medium risk (30% ≤ probability of suicide attempts <60%), and high risk (probability of suicide attempts ≥60%). Medical staff can promptly develop personalized intervention measures for patients based on risk levels, such as arranging close monitoring and emergency psychological intervention for high-risk patients, regular assessment and psychological counseling for medium-risk patients, and providing mental health education materials for low-risk patients.

[0100] Through the above-described specific implementation methods, the prediction methods and systems of the present invention have demonstrated excellent results in practical applications. The rigorous screening and comprehensive evaluation of data collection ensured the homogeneity of the research subjects and the reliability of the data, providing a solid foundation for subsequent analysis and model building. During the data analysis and model building process, a nomogram model constructed by combining multiple methods with the six optimal variables screened was comprehensively evaluated and demonstrated excellent discrimination, calibration, and clinical applicability.

[0101] In a simulation test of an actual clinical scenario, 100 new adolescent depression patients were subjected to suicide risk prediction, of which 30 patients had actually attempted suicide and 70 patients had not. Using the prediction system of the present invention for evaluation, 23 patients who attempted suicide were correctly identified (with a sensitivity of 76.7%, close to the sensitivity of the validation cohort of 57.9%), and 60 patients who did not attempt suicide were correctly identified (with a specificity of 85.7%, close to the specificity of 81.3% of the training cohort and the specificity of 87.5% of the validation cohort). This shows that the system can more accurately judge the suicide risk status of patients, providing strong support for medical staff to promptly identify high-risk patients.

[0102] Furthermore, the prediction system is easy to use and highly efficient, significantly reducing assessment time. Traditional assessment methods require medical staff to spend considerable time researching data and conducting manual calculations and analyses. However, this system requires only data input to rapidly generate risk assessment results, significantly improving clinical efficiency. This helps medical staff implement timely intervention measures, which is crucial for reducing the suicide risk of adolescents with depression.

[0103] This example demonstrates ingenious R&D concepts and outstanding results in assessing and predicting suicide risk in adolescents with depression, including the definition of adolescents, the scope of data collection, and the construction of key predictive factors and nomograms.

[0104] 1. Precise focus on specific groups: Adolescents are defined as those aged 13 to 20, and research is conducted specifically on patients with depression in this age group, which is different from other broad studies on adolescent suicide risk. This precise positioning fully takes into account the uniqueness of patients with depression in this age group in terms of physiology, psychological development, and social environment influences, providing the possibility of more accurate assessment of suicide risk. For example, adolescents aged 13-20 are in a critical period of rapid physical and mental development and social role transformation. The manifestations of depression and suicide risk factors at this stage are significantly different from those in other age groups. This research method of precisely focusing on specific groups is an innovative breakthrough in existing technologies.

[0105] 2. Screening key predictors: The six best predictors of suicide attempts (HAMD, PHQ-15, BSSI, emotional neglect, physical neglect, and low extraversion) were selected from a large number of potential factors. These factors comprehensively cover multiple key dimensions, including clinical symptoms (HAMD, PHQ-15), suicidal ideation (BSSI), childhood trauma (emotional neglect, physical neglect), and personality traits (low extraversion). Unlike other studies that may only focus on risk factors of a single or a few dimensions, the multi-dimensional screening method of this patent more comprehensively captures the complex factors that affect suicide risk and improves the accuracy and comprehensiveness of the prediction model. This multi-dimensional and precise method of screening predictive factors reflects the creativity of this patent in terms of technical solutions.

[0106] 3. Construct an innovative nomogram model: Based on the six key predictive factors screened out, a nomogram is constructed to assess suicide risk. The nomogram transforms complex multi-factor models into intuitive graphics, making it easier for medical staff and researchers to use. By assigning scores to each predictive factor and aggregating the total scores, the suicide risk probability is intuitively reflected, greatly improving the operability and visualization of risk assessment. Compared with traditional risk assessment models, the nomogram model has significant innovations in presentation and ease of use, providing a more efficient and intuitive tool for suicide risk assessment.

[0107] 4. Breakthrough in traditional research ideas: Traditional research on adolescent suicide risk is often broad in scope, and does not fully consider the impact of depression, a specific disease, on suicide risk. It is also rare to comprehensively screen predictive factors from multiple dimensions and build intuitive models. Before this patent, there was a certain inertia in related field research. Most studies conducted suicide risk assessments on a broad group of adolescents, and did not focus in depth on specific adolescents with depression. Influenced by this traditional research model, those skilled in the art are accustomed to studying adolescent suicide issues from a macro perspective, and pay less attention to the in-depth impact of disease factors on suicide risk. For example, past studies have mostly focused on the common characteristics and social environment of adolescents, and have ignored the specific manifestations of depression as an important predisposing factor for suicide in different age groups. This inertial thinking makes it difficult for technicians to break through the inherent framework and think of precisely positioning the research subjects on depressed adolescents aged 13-20, and then deeply explore related suicide risk factors.

[0108] This patent breaks with this traditional approach by precisely targeting patients aged 13-20 with depression, deeply exploring multidimensional risk factors, and innovatively constructing a nomogram model. This research direction and technical solution are not readily apparent to those skilled in the art based on existing technology. It requires a deep understanding and innovative thinking about the pathological mechanisms of adolescent depression, suicide risk factors, and model construction methods.

[0109] 5. Overcome the difficulty of technical integration and solve practical application problems: This embodiment comprehensively uses a variety of professional knowledge and technical means to achieve multi-dimensional data collection, complex statistical analysis and innovative model construction. From the perspective of data collection, it involves the comprehensive use of multiple professional scales, covering multiple fields such as mental diagnosis, suicidal behavior assessment, personality trait measurement and childhood trauma assessment. Those skilled in the art are usually only familiar with the assessment tools in their own professional fields, and it is difficult to organically integrate and apply scales in these different fields. In terms of analysis methods, combining univariate logistic regression, random forest and lasso regression to screen variables requires technicians to have solid statistical knowledge and rich practical experience, and to be proficient in the principles and application scenarios of various analysis methods. Not all technicians can do it easily. Constructing a nomogram model and using multiple methods for evaluation also places high demands on the cross-domain knowledge integration ability of technicians. The complexity of this multi-domain technology integration exceeds the conventional thinking and ability of ordinary technicians.

[0110] In actual clinical applications, an accurate, convenient, and visual suicide risk assessment tool is needed. This patent successfully addresses this challenge through its unique design. Its precise population targeting, screening of key predictive factors, and construction of a nomogram model effectively improve the accuracy and operability of suicide risk assessment. This effective solution, tailored to practical application needs, is not simply an improvement on existing technologies, but is achieved through in-depth research and innovative thinking, demonstrating the non-obviousness of this patent.

[0111] 6. Breaking through the prejudice of insufficient understanding of the relationship between disease and suicide risk, and bringing significant technical effects: Although the association between depression and suicide risk has been recognized to a certain extent, technicians in this field lack in-depth research on the special influencing factors of suicide risk in patients with depression aged 13-20. Adolescents in this age group are in a critical period of rapid physical and mental development and social role transformation. The manifestations of depression and suicide risk factors at this stage are significantly different from those in other age groups. Technicians often ignore these differences and fail to deeply explore suicide risk prediction factors suitable for this specific group. For example, the personality traits of adolescents are still unstable at this stage, and the relationship between low extroversion and suicide risk may be different from that of adults, but existing studies have paid little attention to this. Due to the lack of in-depth understanding of this special relationship, technicians find it difficult to think of incorporating specific factors such as low extroversion into the suicide risk prediction model.

[0112] The implementation of this embodiment has brought significant technical effects. For example, in actual clinical scenario simulation tests, it demonstrated good discrimination, calibration, and clinical applicability. Its prediction model achieved high area under the curve (AUC) values ​​in both the training cohort and the validation cohort, and was able to accurately distinguish between patients who attempted suicide and those who did not attempt suicide. This significant improvement in technical effects further demonstrates the non-obviousness of this embodiment relative to the prior art, indicating that it is a technical solution with outstanding substantive features obtained through creative work.

[0113] Example 2

[0114] Different from Example 1, the method for predicting suicide risk of adolescent depression in this embodiment, in addition to using existing assessment tools, also collects the adolescent's Life Event Stress Scale (LES) scores in the past year in the data collection step to assess the impact of life events on their psychological state; at the same time, the adolescent's sleep quality assessment data is collected and quantitatively assessed using the Pittsburgh Sleep Quality Index (PSQI).

[0115] Adding Life Event Stress Scale scores and sleep quality assessment data provides a more comprehensive understanding of the psychological stressors and sleep patterns of adolescents with depression. Life events are often a significant factor in triggering suicide risk, and sleep quality is closely related to depression and suicide risk. By collecting this data, we can uncover more potential suicide risk factors and improve the accuracy and comprehensiveness of the prediction model.

[0116] In specific implementation, when conducting research or applying this prediction method, adolescents with depression who meet the study criteria will be instructed by trained professionals to complete the Life Event Stress Scale (LES) over the past year. This scale covers common life events in family, work, study, social life, and other areas. Patients are asked to tick the events they experienced in the past year based on their individual circumstances and self-rate the level of stress caused by each event. The scale is divided into five levels (0, 1, 2, 3, and 4 points, respectively) ranging from no impact to extremely impactful. After the collection of information, the total LES score is calculated according to the scale's scoring rules to quantify the impact of life events on the patient's psychological state.

[0117] At the same time, the Pittsburgh Sleep Quality Index (PSQI) is used to assess patients' sleep quality. The PSQI consists of seven dimensions: sleep quality, time to sleep onset, sleep duration, sleep efficiency, sleep disorders, hypnotic medication use, and daytime dysfunction, with each dimension scored from 0 to 3 points. Professionals guide patients to fill out a questionnaire based on their sleep status over the past month. Upon completion, the total PSQI score is calculated, ranging from 0 to 21 points, with higher scores indicating worse sleep quality.

[0118] The LES score, as an indicator of the degree of stress quantified by life events, varies depending on individual experiences. However, higher total scores indicate a greater potential negative impact of life events on adolescents' psychological well-being, and a corresponding increase in suicide risk. The PSQI has a total score of 0-21. Generally, a PSQI score greater than 7 indicates sleep quality problems. Higher scores indicate more severe sleep problems and a stronger association with adolescent depression and suicide risk. Including scores on these two scales in data collection provides richer quantitative data for subsequent analysis, facilitating more accurate predictions of suicide risk.

[0119] Example 3

[0120] Different from Example 1, in the method for predicting suicide risk of adolescent depression in this embodiment, in the feature selection step, principal component analysis (PCA) is used to reduce the dimensionality of the screened variables to reduce redundant information between variables; in the model evaluation step, in addition to the existing evaluation method, the Bootstrap resampling method is also used to internally verify the model to further evaluate the stability of the model.

[0121] Principal component analysis can simplify data structures, reduce model complexity, avoid overfitting caused by too many dependent variables, and improve model efficiency and generalization. The bootstrap resampling method, through repeated sampling and modeling, can more accurately assess the stability and reliability of the model, providing a stronger foundation for its practical application.

[0122] In specific implementation, after feature selection is completed, multiple variables such as the Hamilton Depression Rating Scale (HAMD) score and the Patient Health Questionnaire Somatic Symptom Cluster Scale (PHQ-15) score are obtained. When using the principal component analysis (PCA) method, the data of these variables are first standardized to eliminate the dimension effect. Then the correlation coefficient matrix of the variables is calculated, and the eigenvalues ​​and eigenvectors are obtained by solving the characteristic equation. The number of principal components is determined according to the cumulative contribution rate, and generally the principal components with a cumulative contribution rate of 80%-90% are selected. The original variables are converted into principal components to achieve the purpose of dimensionality reduction, reduce the redundant information between variables, and make the subsequent model construction more concise and effective.

[0123] When using the bootstrap resampling method during model evaluation, repeated sampling with replacement is performed from the original training dataset, each time extracting data of the same size as the original sample, to generate multiple bootstrap samples. A nomogram model is constructed for each bootstrap sample, and corresponding evaluation metrics such as the area under the curve (AUC), specificity, and sensitivity are calculated. After repeating the sampling and modeling process multiple times (e.g., 1000 times), the distribution of these evaluation metrics is analyzed to assess model stability.

[0124] In principal component analysis, the number of principal components is determined by calculating eigenvalues ​​and cumulative contribution rates, thereby achieving dimensionality reduction of the original variables. For example, if, after analysis, the cumulative contribution rate of the first three principal components reaches 85%, these three principal components are selected to replace the original multiple variables for subsequent modeling. In bootstrap resampling, after repeated sampling to build the model, statistics such as the mean and standard deviation of the evaluation indicators are calculated. Taking AUC as an example, if the mean AUC obtained after 1000 resamplings is 0.82 and the standard deviation is 0.03, it indicates that the model has good average discriminatory ability, and the AUC value is relatively stable with small fluctuations, further proving the strong stability of the model. These quantitative results provide a more reliable basis for evaluating the performance and stability of the model.

[0125] Example 4

[0126] Different from the first embodiment, the risk assessment module of the adolescent depression suicide risk prediction system in this embodiment also has a risk trend analysis function. Based on the patient's multiple assessment data, a time series analysis algorithm is used to predict the changing trend of suicide risk. At the same time, the system is equipped with an intervention measure recommendation module, which intelligently recommends personalized intervention measures, such as psychological treatment plans, drug treatment suggestions, etc., according to the risk level and individual characteristics of the patient.

[0127] The risk trend analysis function helps medical staff promptly identify dynamic changes in patients' suicide risk and prepare for intervention in advance. The intervention recommendation module provides medical staff with targeted intervention suggestions, improving the accuracy and effectiveness of interventions, better meeting the individual needs of patients, and more effectively reducing the suicide risk of adolescents with depression.

[0128] During the specific implementation, when the risk trend analysis function of the risk assessment module is realized, the system automatically collects assessment data of the same patient at different time points, including but not limited to key indicators such as the Hamilton Depression Rating Scale (HAMD) score and the Beck Suicide Ideation Inventory (BSSI) score. Time series analysis algorithms, such as the autoregressive integrated moving average model (ARIMA), are used to model and analyze these data. First, the data is tested for stationarity. If the data is not stationary, differential processing is performed to make it stationary. Then, the order of the model, such as p, d, and q values, is determined based on the autocorrelation function (ACF) and partial autocorrelation function (PACF). By fitting the model and performing parameter estimation, a prediction model is obtained. The model is used to predict the patient's suicide risk indicators in the future, thereby analyzing the changing trend of suicide risk.

[0129] The intervention recommendation module makes intelligent recommendations based on the risk level (e.g., low risk, medium risk, high risk) derived from the risk assessment module and the patient's individual characteristics (e.g., age, gender, personality traits, disease severity, etc.). An intervention database is established, containing the applicable scope of different psychotherapy programs (cognitive behavioral therapy, dialectical behavioral therapy, etc.) and drug treatment recommendations (applicable symptoms and dosage ranges for different drugs, etc.). Once the system obtains the patient's risk level and individual characteristics, it matches the corresponding intervention measures in the database through a preset recommendation algorithm and displays them in order of matching.

[0130] In terms of risk trend analysis, the time series analysis model is used to predict the predicted value of the patient's suicide risk index at a specific time point in the future (such as the next 1 month or 3 months) to quantify the changing trend of suicide risk. For example, it is predicted that a patient's HAMD score may rise from the current 20 points to 25 points in 1 month, indicating that his suicide risk is on an upward trend. In the intervention recommendation module, the accuracy and personalization of the recommendation can be quantified by calculating the matching degree between the recommended intervention and the patient's characteristics. The matching degree can be set and calculated based on factors such as feature similarity and clinical experience. For example, the matching degree is expressed as a value of 0-1. The closer the value is to 1, the more the recommended intervention matches the patient's individual characteristics, providing a more targeted reference for medical staff.

[0131] The above is only an embodiment of the present invention, and the common knowledge such as the specific technical solutions and / or characteristics in the solution are not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the description can be used to interpret the content of the claims.

Claims

1. A method for predicting suicide risk in adolescents with depression, characterized in that: The following steps are involved: Step 1: Identify the research subjects. Adolescents aged 13 to 20 years, diagnosed with major depression, not taking medication for at least 3 months, with a total score of ≥17 on the 17-item Hamilton Depression Rating Scale and ≥40 on the Children's Depression Rating Scale-Revised were selected. Individuals with severe physical illness, a history of epileptic seizures and organic brain diseases, schizophrenia, and psychoactive substance abuse were excluded. Step 2: Using an interview questionnaire and the Columbia-Suicide Severity Rating Scale to assess suicidal thoughts and behaviors, participants were divided into a suicide attempt group and a non-suicide attempt group. The Big Five Personality Inventory was used to assess Big Five personality traits, and the Childhood Trauma Questionnaire was used to assess childhood abuse. Step 3: Participants were randomly divided into a training cohort and a validation cohort in a 7:3 ratio. Descriptive analyses were performed using R software. Statistical analysis was performed using t-tests, rank-sum tests, and chi-square tests, combined with univariate logistic regression, random forest, and lasso regression models to identify six optimal predictors of suicide. Step 4: Construct a suicide attempt risk prediction nomogram using the six optimal variables. Assign a preliminary graphic score from 0 to 100 to each predictor. The scores are then determined along the predictor line and summed to obtain a total score. The total score corresponds to the "total score" axis, and the bottom scale reflects the probability of suicide attempt. Step 5: The receiver operating characteristic curve, area under the ROC curve, calibration curve, Hosmer-Lemeshow test, and decision curve analysis were used to evaluate the predictive accuracy, goodness of fit, clinical applicability, and generalization ability of the nomogram model.

2. The method for predicting suicide risk of adolescent depression according to claim 1, characterized in that: The univariate logistic regression was performed on all samples in the training cohort to screen out variables that were significantly associated with suicide attempts and visualized using a forest plot.

3. The method for predicting suicide risk of adolescent depression according to claim 1, characterized in that: The random forest analysis reflects the contribution of each independent variable to the level of suicide attempts by generating an average Gini coefficient reduction. The top ten variables are selected in descending order of contribution value. During the analysis, it is determined that the estimated error rate of out-of-bag samples is lowest when the number of decision trees is 800 and the number of variables is 4.

4. The method for predicting suicide risk of adolescent depression according to claim 1, characterized in that: The lasso regression method selects optimal parameters through ten-fold cross validation, screens predictive factors of suicide attempts from 22 potential factors, reduces the correlation between variables, and avoids model overfitting.

5. The method for predicting suicide risk of adolescent depression according to claim 1, characterized in that: The six optimal variables are Hamilton Depression Rating Scale score, Patient Health Questionnaire Somatic Symptom Cluster Scale score, Beck Suicidal Ideation Scale score, extraversion, emotional neglect, and physical neglect.

6. A suicide risk prediction system for adolescent depression, characterized by: The method for predicting suicide risk of adolescent depression according to claim 1 comprises: The data collection module is used to collect relevant data of adolescent patients with depression, including basic information, diagnostic information, and various scale assessment data; The data analysis module uses a variety of statistical analysis methods to process and analyze the collected data and screen out key predictive factors; Model construction module, which constructs a suicide risk prediction nomogram model based on the screened predictive factors and optimizes and adjusts the model; The risk assessment module substitutes the patient's specific data into the nomogram model to calculate the probability of suicide attempt and assess the patient's suicide risk level.

7. The suicide risk prediction system for adolescent depression according to claim 6, characterized in that: The data acquisition module is designed with a special data entry interface, which has the functions of prompting required items and checking data format.

8. The suicide risk prediction system for adolescent depression according to claim 6, characterized in that: The data analysis module integrates the analysis functions of SPSS and R software to realize an automated analysis process.

9. The suicide risk prediction system for adolescent depression according to claim 6, characterized in that: The model building module uses professional drawing tools to build a nomogram model and provides model parameter adjustment and optimization functions.

10. The suicide risk prediction system for adolescent depression according to claim 6, characterized in that: The risk assessment module displays risk levels in an intuitive manner, including low risk, medium risk, and high risk.

Images