Out-of-hospital intelligent comprehensive intervention management method and system for depression

Abstract

The invention discloses an out-of-hospital intelligent comprehensive intervention management method and system for depression, and the method comprises the steps: collecting the in-hospital clinical data of a depression patient in a hospital and the out-of-hospital monitoring data of the depression patient outside the hospital, and processing the in-hospital clinical data and the out-of-hospital monitoring data into a subsequent data format for feature engineering and model training; integrating hospital internal clinical data and hospital external monitoring data, and processing the hospital internal clinical data and the hospital external monitoring data into input variables of a comprehensive intervention early warning model based on domain knowledge and feature engineering; and training the comprehensive intervention early warning model based on a multilayer perceptron neural network, and outputting early warning management level data of multiple dimensions including safety risk management, drug safety management, medication compliance management, health propaganda and education management and lifestyle management. According to the invention, the effectiveness and safety of depression treatment are improved, and more convenient and timely support is provided for patients.

Description

Technical Field

[0001] The present invention relates to the field of medical and health technologies, and in particular discloses an intelligent comprehensive intervention management method and system for depression outside the hospital. Background Art

[0002] Existing methods for the treatment and intervention of depression mostly focus on hospital treatment and drug treatment, lacking effective monitoring and intervention for the condition of patients outside the hospital. Although digital tools have been used for out-of-hospital monitoring, these tools often cannot be effectively combined with the clinical data of patients, resulting in the lack of personalization and comprehensiveness in treatment intervention.

[0003] Therefore, the above-mentioned defects existing in the existing methods for the treatment and intervention of depression are technical problems that need to be solved urgently at present. Summary of the Invention

[0004] The present invention provides an intelligent comprehensive intervention management method and system for depression outside the hospital, aiming to solve the above-mentioned defects existing in the existing methods for the treatment and intervention of depression.

[0005] One aspect of the present invention relates to an intelligent comprehensive intervention management method for depression outside the hospital, including the following steps:

[0006] Respectively collect the in-hospital clinical data of depression patients in the hospital and the out-of-hospital monitoring data outside the hospital, and process the in-hospital clinical data and the out-of-hospital monitoring data into a data format for subsequent feature engineering and model training; among them, the in-hospital clinical data includes test and examination data and clinical symptom data; the out-of-hospital monitoring data includes the physiological index monitoring data, daily behavior data, medication data, psychological training and science popularization and education data of patients outside the hospital;

[0007] Integrate the in-hospital clinical data and the out-of-hospital monitoring data, and based on domain knowledge and feature engineering, process the in-hospital clinical data and the out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model;

[0008] Train the comprehensive intervention early warning model based on a multi-layer perceptron neural network, and output early warning management level data including multiple dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management;

[0009] Define the management content and specific management measures of the comprehensive intervention based on the different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management;

[0010] Based on the management content and specific management measures of the comprehensive intervention, form system automated execution tasks, and send the specific intervention tasks and comprehensive intervention management tasks to the medical staff side or the patient side;

[0011] The medical care side receives specific intervention tasks and manages the patient side; meanwhile, it adjusts the personal information, medical history information, and mental health assessment information in the clinical records based on the changes in the condition of patients with depression and the situation of follow-up; and evaluates the management situation, and adjusts the relevant parameters of the comprehensive intervention early warning model based on the evaluation results;

[0012] The patient side receives comprehensive intervention management tasks, records daily life data, and interacts with the medical care side;

[0013] Evaluate the intervention situation of patients with depression from five dimensions: safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjust the risk level and intervention plan as the basis for subsequent training and management;

[0014] The comprehensive intervention management model includes an input layer, a first hidden layer, a second hidden layer, and an output layer,

[0015] The input layer, including 208 neurons, receives a 208-dimensional input feature vector, X = [P, H, L, U, PH, M, MH, B, BC, TH, SH];

[0016] The first hidden layer, connected to the input layer, includes 128 neurons, H1 = ReLU(W1 * X + b1);

[0017] The second hidden layer, connected to the first hidden layer, includes 100 neurons, H2 = ReLU(W2 * H1 + b2);

[0018] The output layer, connected to the second hidden layer, includes 15 neurons, corresponding to different intervention types and intervention levels respectively, Y = Sigmoid(W3 * H2 + b3);

[0019] Among them, X represents the input layer, H1 represents the first hidden layer, H2 represents the second hidden layer, Y represents the output layer, P represents the personal basic information feature vector, H represents the personal medical history and family medical history feature vector, L represents the lifestyle feature vector, U represents the medication data feature vector, PH represents the physical examination feature vector, M represents the mental status examination feature vector, MH represents the mental health feature vector, B represents the routine biochemical feature vector, BC represents the routine biochemical feature vector, TH represents the thyroid function feature vector, and SH represents the sex hormone combination feature vector; W1, W2, and W3 are the weight matrices of the first hidden layer, the second hidden layer, and the output layer respectively. The value range of W1 is [-0.01, 0.01], the value range of W2 is [-0.15, 0.15], and the value range of W3 is [-0.02, 0.02]. b1, b2, and b3 are the bias vectors of the first hidden layer, the second hidden layer, and the output layer respectively, and the value range is [0, 0.01]; ReLU is the activation function, and Sigmoid is the activation function of the output layer.

[0020] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention warning model based on domain knowledge and feature engineering, the comprehensive intervention warning model is a multi-task deep learning model based on a multi-layer perceptron neural network, with seven categories of data features including personal basic information variables, personal medical history and family medical history variables, lifestyle variables, medication data variables, clinical observation and evaluation data variables, mental health status data variables, and physiological index data variables as input variables, and five dimensions including safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management as output variables.

[0021] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the feature parameters, value range, and data processing methods of the personal basic information variables. The domain knowledge and feature engineering include personal basic information features. The personal basic information feature vector is a 14-dimensional vector composed of features related to age, gender, economic status, living environment, and education level:

[0022] P = [A_norm, G_male, G_female, G_other, E_low, E_medium, E_high, L_urban, L_suburban, L_rura, Ed_uneducated, Ed_primary, Ed_secondary, Ed_university]

[0023] Among them, P represents the personal basic information feature vector; A_norm is the value after normalizing the patient's age; G_male, G_female, and G_other represent the categorical variables of male, female, and others respectively; E_low, E_medium, and E_high represent the sample economic level data, which are divided into three categories: low income, medium income, and high income; L_urban, L_suburban, and L_rural are the living environment data, representing urban, suburban, and rural respectively; Ed_uneducated, Ed_primary, Ed_secondary, and Ed_uneducated are the education level variables, corresponding to uneducated, primary school, middle school, and university respectively.

[0024] A_norm = (A - A_min) / (A_max - A_min)

[0025] Among them, A_norm is the value after normalizing the patient's age; A_min and A_max represent the minimum and maximum values of the patient's age respectively; A represents the average age of the patient.

[0026] Domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of personal medical history and family medical history. Domain knowledge and feature engineering include the personal medical history and family medical history feature vectors. The personal medical history and family medical history feature vectors are 12-dimensional vectors composed of features related to the history of past chronic diseases, the history of past infectious diseases, the number of family members with depression, and the number of family members with mental illness:

[0027] H = [D_diabetes, D_hypertension, D_hear, D_copd, D_cancer, I_flu, I_tb, I_hepatitis, I_syphilis, I_hiv, F_depression_norm, F_mental_norm]

[0028] Among them, H represents the personal medical history and family medical history feature vector. D_diabetes, D_hypertension, D_heart, D_copd, and D_cancer represent the common types of past chronic diseases such as diabetes, hypertension, heart disease, chronic obstructive pulmonary disease (COPD), and cancer after one-hot encoding respectively; I_flu, I_tb, I_hepatitis, I_syphilis, and I_hiv represent the common types of past infectious diseases such as influenza, tuberculosis, hepatitis, syphilis, and AIDS after one-hot encoding respectively; F_depression_norm and F_mental_norm represent the number of family members with depression and the number of family members with mental illness after normalization respectively.

[0029] Domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of lifestyle variables. Domain knowledge and feature engineering include lifestyle feature vectors, and the lifestyle feature vector is a 13-dimensional vector composed of several related features such as eating habits, smoking habits, drinking habits, exercise habits, and sleep conditions:

[0030] L = [D_normal, D_meat, D_veg, D_light, D_heavy, S, R_light, R_heavy, R_none, C_norm, T_norm, D_s_norm, N_norm]

[0031] Among them, L represents the lifestyle feature vector. D_normal, D_meat, D_veg, D_light, and D_heavy are the classification information of eating habits, corresponding to normal eating habits, meat-eating habits, vegetable-eating habits, light eating habits, and heavy eating habits respectively; S is the classification information of smoking habits; R_light, R_heavy, and R_none are the classification information of drinking habits, corresponding to light drinking habits, heavy drinking habits, and no drinking habits respectively; C is the daily exercise step information after normalization; T_norm, D_s_norm, and N_norm represent the sleep-related feature information after normalization, which are mild insomnia, moderate sleep, and severe insomnia respectively.

[0032] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of commonly used drugs for depression patients. Domain knowledge and feature engineering include medication data feature vectors, and the medication data feature vector is a 24-dimensional vector composed of related features such as drug names, drug doses, days of taking, and frequencies of taking:

[0033] U = [M1, M2, M3, M4, M5, M6, D_{M1}, D_{M2}, D_{M3}, D_{M4}, D_{M5}, D_{M6}, T_{M1}, T_{M2}, T_{M3}, T_{M4}, T_{M5}, T_{M6}, F_{M1}, F_{M2}, F_{M3}, F_{M4}, F_{M5}, F_{M6}]

[0034] Among them, U represents the medication data feature vector, M1, M2, M3, M4, M5, and M6 respectively represent 6 kinds of drugs; D_{M1}, D_{M2}, D_{M3}, D_{M4}, D_{M5}, and D_{M6} respectively represent the dosage of each corresponding drug; T_{M1}, T_{M2}, T_{M3}, T_{M4}, T_{M5}, and T_{M6} respectively represent the number of days of taking each corresponding drug, and F_{M1}, F_{M2}, F_{M3}, F_{M4}, F_{M5}, and F_{M6} respectively represent the frequency of taking each corresponding drug.

[0035] Domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of physical examination variables. Domain knowledge and feature engineering include a physical examination feature vector, which is a 30-dimensional vector composed of features related to head and facial signs, cardiovascular signs, chest and abdominal signs, and nervous system signs:

[0036] PH = [FS_1, FS_2,..., FS_8, CVS_1, CVS_2,..., CVS_5, TAS_1, TAS_2,.., TAS_5, NS_1, NS_2,..., NS_12]

[0037] Among them, PH represents the physical examination feature vector, FS_1, FS_2,..., FS_8 respectively represent head and facial signs, including premature aging, mydriasis, miosis, nystagmus, abnormal eye movement, abnormal light reflex, abnormal convergence reflex, and thyroid enlargement; CVS_1, CVS_2,..., CVS_5 respectively represent cardiovascular signs, including elevated blood pressure, decreased blood pressure, tachycardia, bradycardia, and abnormal heart sounds; TAS_1, TAS_2,..., TAS_5 represent chest and abdominal signs, including respiratory distress, spider nevus, hepatomegaly, splenomegaly, and ascites; NS_1, NS_2,..., NS_12 represent nervous system signs, including low muscle strength, high muscle strength, hypotonia, hypertonia, hyperreflexia, pathological reflex, meningeal irritation sign, hypoalgesia or absence of superficial reflex, hypoalgesia or absence of deep reflex, sensory disturbance, hemiplegia, and paraplegia.

[0038] Further, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of the mental status examination variables. The domain knowledge and feature engineering include a mental status examination feature vector, which is a 71-dimensional vector composed of features related to perceptual disorders, formal thought disorders, thought content disorders, logical thought disorders, attention disorders, memory disorders, intellectual disorders, disorientation, emotional activities, volitional disorders, and lack of insight:

[0039] M = [PD_1, PD_2,..., PD_6, TPD_1, TPD_2,..., TPD_13, TCD_1, TCD_2,..., TCD_11, LTD_1, LTD_2,..., LTD_6, AD_1, AD_2,..., AD_6, MD_1, MD_2,..., MD_6, ID_1,.., ID_4, OD_1, OD_2, OD_3, EA_1, EA_2,..., EA_9, VD_1, VD_2,..., VD_5, LI_1, LI_2]

[0040] Among them, M represents the mental status examination feature vector, and PD_1, PD_2, ..., PD_6 respectively represent perceptual disorders, including none, auditory hallucination, visual hallucination, olfactory hallucination, gustatory hallucination, and tactile hallucination; TPD_1, TPD_2, ..., TPD_13 respectively represent formal thought disorders, including flight of ideas, thought retardation, poverty of thought, loose association, fragmented thought, incoherence of thought, thought block, thought insertion, thought pressure, circumstantiality, pathological symbolic thinking, neologism, and paralogia; TCD_1, TCD_2, ..., TCD_11 respectively represent disorders of thought content, including delusion of persecution, delusion of reference, delusion of being controlled, delusion of physical influence, delusion of grandeur, delusion of guilt, delusion of jealousy, delusion of love, delusion of non-biological origin, thought echo, and feeling of being revealed; LTD_1, LTD_2, ..., LTD_6 respectively represent disorders of thought logic, including pathological symbolic thinking, neologism, symbolic thinking, illogical thinking, paralogia, and sophism; AD_1, AD_2, ..., AD_6 respectively represent attention disorders, including hyperprosexia, hypoprosexia, circumstantiality of attention, obtention of attention, narrowing of attention, and fixation of attention; MD_1, MD_2, ..., MD_6 respectively represent memory disorders, including hypomnesia, hypermnesia, amnesia, confabulation, fabrication, and cryptomnesia; ID_1, ..., ID_4 respectively represent intellectual disorders, including mild, moderate, severe, and profound mental retardation; OD_1, OD_2, ..., OD_3 respectively represent disorders of orientation, including disorders of person orientation, place orientation, and time orientation; EA_1, EA_2, ..., EA_9 respectively represent emotional activities, including elation, depression, anxiety, fear, emotional flatness, irritability, emotional fragility, emotional paraphrenia, and ambivalence; VD_1, VD_2, ..., VD_5 respectively represent volitional disorders, including hypobulia, hyperbulia, psychomotor excitement, psychomotor inhibition, and strange actions or behaviors; LI_1 and LI_2 respectively represent the classification of lack of insight, 1 represents lack, and 0 represents no lack.

[0041] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of mental health status variables. The domain knowledge and feature engineering include a mental health feature vector, which is an 8-dimensional vector composed of the Young Mania Rating Scale score and the Hamilton Depression Rating Scale score:

[0042] MH = [YMRS_Normal, YMRS_Mild, YMRS_Moderate, YMRS_Severe, YMRS_VerySevere, HAMD_NoDepression, HAMD_MildToModerate, HAMD_Severe]

[0043] Among them, MH represents the mental health feature vector, and YMRS_Normal, YMRS_Mild, YMRS_Moderate, YMRS_Severe, YMRS_VerySevere represent the Young Mania Rating Scale scores, representing normal, mild mania, moderate mania, severe mania, and very severe mania respectively; HAMD_NoDepression, HAMD_MildToModerate, HAMD_Severe represent the Hamilton Depression Rating Scale scores, representing no depression, mild to moderate depression, and severe depression respectively.

[0044] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of the physiological index data - blood routine variables. The domain knowledge and feature engineering include a blood routine feature vector, which is a 9-dimensional vector composed of red blood cell count, hemoglobin, white blood cell count, platelet count, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin content, mean corpuscular hemoglobin concentration, and red blood cell distribution width:

[0045] B = [RBC_norm, Hb_norm, WBC_norm, Platelets_norm, HCT_norm, MCV_norm, MCH_norm, MCHC_norm, RDW_norm]

[0046] Among them, B represents the blood routine feature vector, RBC_norm represents the red blood cell count, Hb_norm represents the hemoglobin count, WBC_norm represents the white blood cell count, Platelets_norm represents the platelet count, HCT_norm represents the hematocrit, MCV_norm represents the mean corpuscular volume, MCH_norm represents the mean corpuscular hemoglobin content, MCHC_norm represents the mean corpuscular hemoglobin concentration, and RDW_norm represents the red blood cell distribution width;

[0047] RBC_norm = (RBC - RBC_min) / (RBC_max - RBC_min)

[0048] Among them, RBC_norm represents the red blood cell count, RBC_max and RBC_min are the maximum and minimum values of the red blood cell count respectively, and RBC represents the average value of the red blood cell count.

[0049] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of physiological index data - biochemical routine variables. The domain knowledge and feature engineering include a biochemical routine feature vector, which is a 15-dimensional vector composed of alanine aminotransferase, aspartate aminotransferase, total bilirubin, serum creatinine, blood urea nitrogen, uric acid, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, triglycerides, fasting blood glucose, sodium electrolyte, potassium electrolyte, calcium electrolyte, and chloride electrolyte:

[0050] BC = [ALT_norm, AST_norm, TBIL_norm, Creatinine_norm, BUN_norm, UricAcid_norm, TotalCholesterol_norm, LDL-C_norm, HDL-C_norm, Triglycerides_norm, FastingGlucose_norm, Na_norm, K_norm, Ca_norm, Cl_norm]

[0051] Among them, BC represents the biochemical routine feature vector, ALT_norm represents alanine aminotransferase, AST_norm represents aspartate aminotransferase, TBIL_norm represents total bilirubin, Creatinine_norm represents serum creatinine, BUN_norm represents blood urea nitrogen, UricAcid_norm represents uric acid, TotalCholesterol_norm represents total cholesterol, LDL-C_norm represents low-density lipoprotein cholesterol, HDL-C_norm represents high-density lipoprotein cholesterol, Triglycerides_norm represents triglycerides, FastingGlucose_norm represents fasting blood glucose, Na_norm represents sodium electrolyte, K_norm represents potassium electrolyte, Ca_norm represents calcium electrolyte, and Cl_norm represents chloride electrolyte;

[0052] ALT_norm = (ALT - ALT_min) / (ALT_max - ALT_min)

[0053] Among them, ALT_norm represents alanine aminotransferase, ALT_max and ALT_min are the maximum and minimum values of alanine aminotransferase respectively, and ALT represents the average value of alanine aminotransferase;

[0054] Domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of physiological index data - sex hormone level variables. Domain knowledge and feature engineering include sex hormone combined feature vectors. The sex hormone combined feature vector includes a 5-dimensional vector composed of testosterone, estradiol, luteinizing hormone, follicle-stimulating hormone, and progesterone:

[0055] SH = [Testosterone_norm, E2_norm, LH_norm, FSH_norm, Progesterone_norm]

[0056] Among them, SH represents the hormone combined feature vector, Testosterone_norm represents testosterone, E2_norm represents estradiol, LH_norm represents luteinizing hormone, FSH_norm represents follicle-stimulating hormone, and Progesterone_norm represents progesterone.

[0057] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention warning model based on domain knowledge and feature engineering, domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of the output variables of the intelligent comprehensive intervention model for depressive disorders. Domain knowledge and feature engineering include intervention warning feature vectors. The intervention warning feature vector includes safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management:

[0058] IW = [SRM_Tip, SRM_Warning, SRM_Danger, MSM_Tip, MSM_Warning, MSM_Danger, MAM_Tip, MAM_Warning, MAM_Danger, HEM_Tip, HEM_Warning, HEM_Danger, LSM_Tip, LSM_Warning, LSM_Danger]

[0059] Among them, IW represents the intervention warning feature, and SRM_Tip, SRM_Warning, and SRM_Danger represent the tip, warning, and danger levels of safety risk management respectively;

[0060] MSM_Tip, MSM_Warning, and MSM_Danger respectively represent the tip, warning, and danger levels of drug safety management; MAM_Tip, MAM_Warning, and MAM_Danger respectively represent the tip, warning, and danger levels of medication compliance management; HEM_Tip, HEM_Warning, and HEM_Danger respectively represent the tip, warning, and danger levels of health education management.

[0061] Another aspect of the present invention relates to an intelligent comprehensive intervention management system for out-of-hospital depression, including:

[0062] A clinical data collection module for collecting in-hospital clinical data of depression patients, where the in-hospital clinical data includes test and examination data and clinical symptom data;

[0063] A patient-side data collection module for collecting out-of-hospital monitoring data, where the out-of-hospital monitoring data includes physiological index monitoring data, daily behavior data, medication data, psychological training, and science popularization and education data of the patient outside the hospital;

[0064] A feature processing module for integrating in-hospital clinical data and out-of-hospital monitoring data, and based on domain knowledge and feature engineering, processing the in-hospital clinical data and out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model;

[0065] A comprehensive intervention early warning model for training the comprehensive intervention early warning model based on a multi-layer perceptron neural network, and outputting early warning management level data including multiple dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management;

[0066] A comprehensive intervention management module for defining the management content and specific management measures of comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management;

[0067] An intervention task management module for defining the management content and specific management measures of comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management;

[0068] A medical staff-side module for receiving specific intervention tasks and managing the patient side; at the same time, adjusting personal information, medical history information, and mental health assessment information in the clinical record based on the changes in the condition of depression patients and the situation of follow-up visits; and evaluating the management situation, and adjusting relevant parameters of the comprehensive intervention early warning model based on the evaluation results;

[0069] A patient-side module for receiving comprehensive intervention management tasks, recording daily life data, and interacting with the medical staff side;

[0070] A regular evaluation module is used to evaluate the intervention situation of depression patients from five dimensions: safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjust the risk level and intervention plan as the basis for subsequent training and management.

[0071] The comprehensive intervention management model includes an input layer, a first hidden layer, a second hidden layer, and an output layer.

[0072] The input layer includes 208 neurons and receives a 208-dimensional input feature vector, X = [P, H, L, U, PH, M, MH, B, BC, TH, SH].

[0073] The first hidden layer is connected to the input layer and includes 128 neurons, H1 = ReLU(W1 * X + b1).

[0074] The second hidden layer is connected to the first hidden layer and includes 100 neurons, H2 = ReLU(W2 * H1 + b2).

[0075] The output layer is connected to the second hidden layer and includes 15 neurons, corresponding to different intervention types and levels respectively, Y = Sigmoid(W3 * H2 + b3).

[0076] Among them, X represents the input layer, H1 represents the first hidden layer, H2 represents the second hidden layer, Y represents the output layer, P represents the personal basic information feature vector, H represents the personal medical history and family medical history feature vector, L represents the lifestyle feature vector, U represents the medication data feature vector, PH represents the physical examination feature vector, M represents the mental status examination feature vector, MH represents the mental health feature vector, B represents the biochemical routine feature vector, BC represents the biochemical routine feature vector, TH represents the thyroid function feature vector, SH represents the sex hormone combination feature vector; W1, W2, and W3 are the weight matrices of the first hidden layer, the second hidden layer, and the output layer respectively. The value range of W1 is [-0.01, 0.01], the value range of W2 is [-0.15, 0.15], the value range of W3 is [-0.02, 0.02], b1, b2, and b3 are the bias vectors of the first hidden layer, the second hidden layer, and the output layer respectively, and the value range is [0, 0.01]; ReLU is the activation function, and Sigmoid is the activation function of the output layer.

[0077] The beneficial effects achieved by the present invention are:

[0078] The present invention provides an intelligent comprehensive intervention management method and system for depression outside the hospital. By separately collecting in-hospital clinical data and out-of-hospital monitoring data of depression patients, and processing the in-hospital clinical data and out-of-hospital monitoring data into a data format for subsequent feature engineering and model training; wherein, the in-hospital clinical data includes test and examination data and clinical symptom data; the out-of-hospital monitoring data includes physiological index monitoring data, daily behavior data, medication data, psychological training and science popularization education data of the patient outside the hospital; integrating the in-hospital clinical data and out-of-hospital monitoring data, and based on domain knowledge and feature engineering, processing the in-hospital clinical data and out-of-hospital monitoring data into input variables of a comprehensive intervention early warning model; training the comprehensive intervention early warning model based on a multi-layer perceptron neural network, and outputting early warning management level data including multiple dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; defining the management content and specific management measures of comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; forming system automated execution tasks based on the management content and specific management measures of comprehensive intervention, and sending specific intervention tasks and comprehensive intervention management tasks to the medical staff side or the patient side; the medical staff side receives specific intervention tasks and manages the patient side; at the same time, adjusting personal information, medical history information, and mental health assessment information in the clinical record based on the changes in the condition of depression patients and the situation of follow-up visits; and evaluating the management situation, and adjusting relevant parameters of the comprehensive intervention early warning model based on the evaluation results; the patient side receives comprehensive intervention management tasks, records daily life data, and interacts with the medical staff side; evaluating the intervention situation of depression patients from 5 dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjusting the risk level and intervention plan, as the basis for subsequent training and management. The intelligent comprehensive intervention management method and system for depression outside the hospital provided by the present invention provide a new comprehensive and personalized treatment and monitoring plan for depression patients by integrating in-hospital and out-of-hospital data and applying an advanced neural network model; using the neural network model to process the collected data, analyzing the treatment response, recurrence risk, and drug safety of patients, and formulating a personalized out-of-hospital treatment and monitoring plan according to the model output, including medication guidance, improvement of living habits, mental health education, etc.; the system can monitor data changes in real time, and give early warnings and interventions for high-risk situations such as suicidal tendency and medication risk; providing remote consultation and emergency contact functions to ensure that patients can obtain professional help in a timely manner when encountering high-risk situations; not only improving the effectiveness and safety of depression treatment, but also providing more convenient and timely support for patients. Brief Description of the Drawings

[0079] Figure 1It is a schematic flow chart of the intelligent comprehensive intervention management method for depression outside the hospital of the present invention;

[0080] Figure 2 It is a schematic structural diagram of the comprehensive intervention management model in the present invention;

[0081] Figure 3 It is a functional block diagram of the intelligent comprehensive intervention management system for depression outside the hospital of the present invention. Specific embodiments

[0082] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings of the specification and specific embodiments.

[0083] As Figure 1 and Figure 2 shown, the first embodiment of the present invention proposes an intelligent comprehensive intervention management method for depression outside the hospital, including the following steps:

[0084] Step S100: Collect the in-hospital clinical data and out-of-hospital monitoring data of depression patients respectively, and process the in-hospital clinical data and out-of-hospital monitoring data into a data format for subsequent feature engineering and model training; among them, the in-hospital clinical data includes inspection and examination data and clinical symptom data; the out-of-hospital monitoring data includes the physiological index monitoring data, daily behavior data, medication data, psychological training and popular science education data of the patient outside the hospital.

[0085] The in-hospital clinical data mainly includes collecting personal basic information, personal history and family history, medication data, physical examination data, mental status examination data, mental health status assessment data, blood routine data, biochemical routine data, thyroid function data, sex hormone level data, etc. from the in-hospital HIS (Hospital Information System), electronic medical records, inspection and examination systems, etc., and processing the data into a data format for subsequent feature engineering and model training.

[0086] The out-of-hospital monitoring data mainly includes collecting personal lifestyle data from the patient-side mobile application and wearable devices, and processing it into a data format for subsequent feature engineering and model training.

[0087] Step S200: Integrate the in-hospital clinical data and out-of-hospital monitoring data, and process the in-hospital clinical data and out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering.

[0088] Integrate the data from the clinical data collection module and the patient - side data collection module. Based on domain knowledge and feature engineering, process the data into input variables for a comprehensive intervention warning model based on MLP (Multi - layer perceptron).

[0089] Step S300: Train the comprehensive intervention warning model based on the multi - layer perceptron neural network, and output warning management level data including multiple dimensions such as safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management.

[0090] Train the comprehensive intervention warning model based on the MLP (Multi - layer perceptron) neural network, and output warning management level data including multiple dimensions such as safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management.

[0091] Step S400: Define the management content and specific management measures for comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management.

[0092] Define the management content and specific management measures for comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management. For example, if the drug safety level is identified as a warning, information will be sent to the doctor to prompt whether the patient needs to adjust or stop taking the medicine. If the suicide risk management level is dangerous, a first - level psychological crisis intervention measure will be initiated, and the attending doctor, the patient's family, and the suicide crisis intervention team will be contacted immediately.

[0093] Step S500: Based on the management content and specific management measures of comprehensive intervention, form system - automated execution tasks, and send the specific intervention tasks and comprehensive intervention management tasks to the medical staff side or the patient side.

[0094] Based on the intervention content and management measures of the comprehensive intervention management module, form system - automated execution tasks and send them to the medical staff side or the patient side. For example, if the comprehensive intervention warning model outputs that the patient needs to strengthen the learning of certain aspects of patient education knowledge, according to the corresponding content in the comprehensive intervention management module, the intervention task management module will generate the corresponding content and send it to the patient side. If the comprehensive intervention warning model identifies that the patient needs to strengthen certain aspects of lifestyle management, the intervention task management module will send the corresponding management tasks and tips to the doctor side and the patient side.

[0095] Step S600: The medical care side receives specific intervention tasks and manages the patient side; meanwhile, based on the changes in the condition of the depression patients and the situation of follow-up, it adjusts the personal information, medical history information, and mental health assessment information in the clinical records; and evaluates the management situation, and adjusts the relevant parameters of the comprehensive intervention warning model based on the evaluation results.

[0096] Receive the specific intervention tasks sent by the intervention task management module and manage the patient side; meanwhile, the medical care side module can adjust the personal information, medical history information, mental health assessment information, etc. in the clinical records based on the changes in the patient's condition and the situation of follow-up; at the same time, the medical care side will also evaluate the management situation, and adjust the relevant parameters of the comprehensive intervention warning model based on the evaluation results.

[0097] Step S700: The patient side receives the comprehensive intervention management tasks, records daily life data, and interacts with the medical care side.

[0098] Receive the comprehensive intervention management tasks sent by the intervention task management module, such as reminders for medication compliance, mental health knowledge and psychological training content, collect data such as exercise and sleep obtained from wearable devices, record daily life data such as diet and mood, etc., and can interact with the doctor side.

[0099] Step S800: Evaluate the intervention situation of depression patients from five dimensions: safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjust the risk level and intervention plan, which serve as the basis for subsequent training and management.

[0100] Evaluate the intervention situation of depression patients from five dimensions: safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjust the risk level and the intervention plan, which serve as the basis for subsequent training and management.

[0101] For the out-of-hospital intelligent comprehensive intervention management method for depression proposed in this embodiment, the comprehensive intervention management model includes an input layer, a first hidden layer, a second hidden layer, and an output layer, where,

[0102] The input layer, including 208 neurons, receives a 208-dimensional input feature vector, and the input layer is expressed as:

[0103] X = [P, H, L, U, PH, M, MH, B, BC, TH, SH] (1)

[0104] The first hidden layer, connected to the input layer, includes 128 neurons, and the first hidden layer is expressed as:

[0105] H1 = ReLU(W1 * X + b1) (2)

[0106] The second hidden layer, connected to the first hidden layer, includes 100 neurons and is represented as:

[0107] H2 = ReLU(W2 * H1 + b2) (3)

[0108] The output layer, connected to the second hidden layer, includes 15 neurons corresponding to different intervention types and levels, and is represented as:

[0109] Y = Sigmoid(W3 * H2 + b3) (4)

[0110] Among them, X represents the input layer, H1 represents the first hidden layer, H2 represents the second hidden layer, Y represents the output layer, P represents the personal basic information feature vector, H represents the personal medical history and family medical history feature vector, L represents the lifestyle feature vector, U represents the medication data feature vector, PH represents the physical examination feature vector, M represents the mental status examination feature vector, MH represents the mental health feature vector, B represents the biochemical routine feature vector, BC represents the biochemical routine feature vector, TH represents the thyroid function feature vector, and SH represents the sex hormone combination feature vector; W1, W2, and W3 are the weight matrices of the first hidden layer, the second hidden layer, and the output layer respectively. The value range of W1 is [-0.01, 0.01], the value range of W2 is [-0.15, 0.15], and the value range of W3 is [-0.02, 0.02]. b1, b2, and b3 are the bias vectors of the first hidden layer, the second hidden layer, and the output layer respectively, and the value range is [0, 0.01]; ReLU is the activation function, and Sigmoid is the activation function of the output layer.

[0111] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the comprehensive intervention early warning model is a multi-task deep learning model based on a multi-layer perceptron neural network, with seven major data features including personal basic information variables, personal medical history and family medical history variables, lifestyle variables, medication data variables, clinical observation and evaluation data variables, mental health status data variables, and physiological index data variables as input variables, and five dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management as output variables.

[0112] Furthermore, in the intelligent comprehensive intervention management method for depression outside the hospital proposed in this embodiment, in step S200, domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of personal basic information variables. Domain knowledge and feature engineering include personal basic information features. The personal basic information feature vector is a 14-dimensional vector composed of features related to age, gender, economic status, living environment, and education level:

[0113] P = [A_norm, G_male, G_female, G_other, E_low, E_medium, E_high, L_urban, L_suburban, L_rura, Ed_uneducated, Ed_primary, Ed_secondary, Ed_university] (5)

[0114] Among them, P represents the personal basic information feature vector; A_norm is the value after normalizing the patient's age; G_male, G_female, and G_other respectively represent the categorical variables of male, female, and others. After one-hot encoding, they are converted into three binary variables G_male, G_female, and G_other; E_low, E_medium, and E_high represent the sample economic level data, which are divided into three categories: low income, medium income, and high income. After one-hot encoding, they are converted into three binary variables E_low, E_medium, and E_high; L_urban, L_suburban, and L_rural are the living environment data, representing urban, suburban, and rural respectively; Ed_uneducated, Ed_primary, Ed_secondary, and Ed_university are the education level variables, corresponding to uneducated, primary school, middle school, and university.

[0115] A_norm = (A - A_min) / (A_max - A_min) (6)

[0116] Among them, A_norm is the value after normalizing the patient's age; A_min and A_max respectively represent the minimum and maximum values of the patient's age; A represents the average value of the patient's age.

[0117] Domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of personal medical history and family medical history. Domain knowledge and feature engineering include personal medical history and family medical history feature vectors. The personal medical history and family medical history feature vector is a 12-dimensional vector composed of features related to past chronic disease history, past infectious disease history, the number of family members with depression, and the number of family members with mental illness:

[0118] H = [D_diabetes, D_hypertension, D_hear, D_copd, D_cancer, I_flu, I_tb, I_hepatitis, I_syphilis, I_hiv, F_depression_norm, F_mental_norm](7)

[0119] Among them, H represents the feature vector of personal medical history and family medical history. D_diabetes, D_hypertension, D_heart, D_copd, and D_cancer respectively represent the common types of several previous chronic diseases such as diabetes, hypertension, heart disease, chronic obstructive pulmonary disease, and tumor after one-hot encoding; I_flu, I_tb, I_hepatitis, I_syphilis, and I_hiv respectively represent the common types of several previous infectious diseases such as influenza, tuberculosis, hepatitis, syphilis, and AIDS after one-hot encoding; F_depression_norm and F_mental_norm respectively represent the number of family members with depression and the number of family members with mental diseases after normalization processing.

[0120] Domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of lifestyle variables. Domain knowledge and feature engineering include lifestyle feature vectors. The lifestyle feature vector is a 13-dimensional vector composed of several related features such as eating habits, smoking habits, drinking habits, exercise habits, and sleep conditions:

[0121] L = [D_normal, D_meat, D_veg, D_light, D_heavy, S, R_light, R_heavy, R_none, C_norm, T_norm, D_s_norm, N_norm](8)

[0122] Among them, L represents the lifestyle feature vector. D_normal, D_meat, D_veg, D_light, and D_heavy are respectively the classification information of eating habits; S is the classification information of smoking habits; R_light, R_heavy, and R_none are respectively the classification information of drinking habits; C is the information of the number of daily exercise steps after normalization processing; T_norm, D_s_norm, and N_norm respectively represent the sleep-related feature information after normalization processing.

[0123] Domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of commonly used drugs for patients with depression. Domain knowledge and feature engineering include medication data feature vectors. The medication data feature vector is a 24-dimensional vector composed of related features such as drug names, drug doses, days of taking, and frequencies of taking:

[0124] U = [M1, M2, M3, M4, M5, M6, D_{M1}, D_{M2}, D_{M3}, D_{M4}, D_{M5}, D_{M6}, T_{M1}, T_{M2}, T_{M3}, T_{M4}, T_{M5}, T_{M6}, F_{M1}, F_{M2}, F_{M3}, F_{M4}, F_{M5}, F_{M6}](9)

[0125] Among them, U represents the medication data feature vector, M1, M2, M3, M4, M5, and M6 respectively represent 6 kinds of drugs; D_{M1}, D_{M2}, D_{M3}, D_{M4}, D_{M5}, and D_{M6} respectively represent the dosage corresponding to each drug; T_{M1}, T_{M2}, T_{M3}, T_{M4}, T_{M5}, and T_{M6} respectively represent the number of days of taking the corresponding drug, and F_{M1}, F_{M2}, F_{M3}, F_{M4}, F_{M5}, and F_{M6} respectively represent the frequency of taking the corresponding drug.

[0126] Domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of physical examination variables. Domain knowledge and feature engineering include the physical examination feature vector, and the physical examination feature vector is a 30-dimensional vector composed of features related to head and facial signs, cardiovascular signs, chest and abdominal signs, and neurological signs:

[0127] PH = [FS_1, FS_2,..., FS_8, CVS_1, CVS_2,..., CVS_5, TAS_1, TAS_2,.., TAS_5, NS_1, NS_2,..., NS_12](10)

[0128] Among them, PH represents the physical examination feature vector, FS_1, FS_2,..., FS_8 respectively represent head and facial signs, including premature aging, mydriasis, miosis, nystagmus, abnormal eye movement, abnormal light reflex, abnormal convergence reflex, and thyroid enlargement; CVS_1, CVS_2,..., CVS_5 respectively represent cardiovascular signs, including elevated blood pressure, decreased blood pressure, tachycardia, bradycardia, and abnormal heart sounds; TAS_1, TAS_2,..., TAS_5 represent chest and abdominal signs, including respiratory distress, spider nevus, hepatomegaly, splenomegaly, and ascites; NS_1, NS_2,..., NS_12 represent neurological signs, including low muscle strength, high muscle strength, low muscle tone, high muscle tone, hyperreflexia, pathological reflex, meningeal irritation sign, weakened or absent superficial reflex, weakened or absent deep reflex, sensory disturbance, hemiplegia, and paraplegia.

[0129] Domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of mental status examination variables. Domain knowledge and feature engineering include mental status examination feature vectors. The mental status examination feature vector is a 71-dimensional vector composed of features related to perceptual disorders, formal thought disorders, thought content disorders, logical thought disorders, attention disorders, memory disorders, intellectual disorders, disorientation, emotional activities, volitional disorders, and lack of insight:

[0130] M = [PD_1, PD_2,..., PD_6, TPD_1, TPD_2,..., TPD_13, TCD_1, TCD_2,..., TCD_11, LTD_1, LTD_2,..., LTD_6, AD_1, AD_2,..., AD_6, MD_1, MD_2,..., MD_6, ID_1,.., ID_4, OD_1, OD_2, OD_3, EA_1, EA_2,..., EA_9, VD_1, VD_2,..., VD_5, LI_1, LI_2](11)

[0131] Among them, M represents the mental status examination feature vector, PD_1, PD_2, ..., PD_6 respectively represent perceptual disorders, including none, auditory hallucination, visual hallucination, olfactory hallucination, gustatory hallucination, and tactile hallucination; TPD_1, TPD_2, ..., TPD_13 respectively represent formal thought disorders, including flight of ideas, thought retardation, poverty of thought, loose association, broken thought, incoherence of thought, thought block, thought insertion, thought pressure, circumstantiality, pathological symbolic thinking, neologism, and paralogic thinking; TCD_1, TCD_2, ..., TCD_11 respectively represent disorders of thought content, including delusion of persecution, delusion of reference, delusion of being controlled, delusion of physical influence, delusion of grandeur, delusion of guilt, delusion of jealousy, delusion of love, delusion of non-blood relationship, thought echo, and feeling of being revealed; LTD_1, LTD_2, ..., LTD_6 respectively represent disorders of thought logic, including pathological symbolic thinking, neologism, symbolic thinking, illogical thinking, paralogic thinking, and sophistry; AD_1, AD_2, ..., AD_6 respectively represent attention disorders, including hyperprosexia, hypoprosexia, circumstantiality of attention, obtuseness of attention, narrowing of attention, and fixation of attention; MD_1, MD_2, ..., MD_6 respectively represent memory disorders, including hypomnesia, hypermnesia, amnesia, confabulation, fabrication, and cryptomnesia; ID_1, ..., ID_4 respectively represent intellectual disabilities, including mild, moderate, severe, and profound mental retardation; OD_1, OD_2, ..., OD_3 respectively represent disorders of orientation, including disorders of person orientation, place orientation, and time orientation; EA_1, EA_2, ..., EA_9 respectively represent emotional activities, including elation, depression, anxiety, fear, emotional flatness, irritability, emotional fragility, emotional paraphrenia, and ambivalence; VD_1, VD_2, ..., VD_5 respectively represent disorders of will, including hypobulia, hyperbulia, psychomotor excitement, psychomotor inhibition, and strange actions or behaviors; LI_1 and LI_2 respectively represent the classification of lack of insight, 1 represents lack, and 0 represents non-lack.

[0132] Furthermore, in the step of integrating in-hospital clinical data and out-of-hospital monitoring data and processing the in-hospital clinical data and out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of mental health status variables. The domain knowledge and feature engineering include a mental health feature vector, which is an 8-dimensional vector composed of the Young Mania Rating Scale score and the Hamilton Depression Rating Scale score:

[0133] MH = [YMRS_Normal, YMRS_Mild, YMRS_Moderate, YMRS_Severe, YMRS_VerySevere, HAMD_NoDepression, HAMD_MildToModerate, HAMD_Severe](12)

[0134] Among them, MH represents the mental health feature vector. YMRS_Normal, YMRS_Mild, YMRS_Moderate, YMRS_Severe, and YMRS_VerySevere represent the Young Mania Rating Scale scores, representing normal, mild mania, moderate mania, severe mania, and very severe mania respectively; HAMD_NoDepression, HAMD_MildToModerate, and HAMD_Severe represent the Hamilton Depression Rating Scale scores, representing no depression, mild to moderate depression, and severe depression respectively.

[0135] Domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of physiological index data - blood routine variables. Domain knowledge and feature engineering include the blood routine feature vector, which is a 9-dimensional vector composed of red blood cell count, hemoglobin, white blood cell count, platelet count, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin content, mean corpuscular hemoglobin concentration, and red blood cell distribution width:

[0136] B = [RBC_norm, Hb_norm, WBC_norm, Platelets_norm, HCT_norm, MCV_norm, MCH_norm, MCHC_norm, RDW_norm](13)

[0137] Among them, B represents the blood routine feature vector, RBC_norm represents the red blood cell count, Hb_norm represents the hemoglobin count, WBC_norm represents the white blood cell count, Platelets_norm represents the platelet count, HCT_norm represents the hematocrit, MCV_norm represents the mean corpuscular volume, MCH_norm represents the mean corpuscular hemoglobin content, MCHC_norm represents the mean corpuscular hemoglobin concentration, and RDW_norm represents the red blood cell distribution width.

[0138] RBC_norm = (RBC - RBC_min) / (RBC_max - RBC_min) (14)

[0139] Among them, RBC_norm represents the red blood cell count, RBC_max and RBC_min are the maximum and minimum values of the red blood cell count respectively, and RBC represents the average value of the red blood cell count.

[0140] Domain knowledge and feature engineering include biochemical routine feature vectors. The biochemical routine feature vectors are 15-dimensional vectors composed of alanine aminotransferase, aspartate aminotransferase, total bilirubin, serum creatinine, blood urea nitrogen, uric acid, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, triglycerides, fasting blood glucose, electrolyte sodium, electrolyte potassium, electrolyte calcium, and electrolyte chloride:

[0141] BC = [ALT_norm, AST_norm, TBIL_norm, Creatinine_norm, BUN_norm, UricAcid_norm, TotalCholesterol_norm, LDL-C_norm, HDL-C_norm, Triglycerides_norm, FastingGlucose_norm, Na_norm, K_norm, Ca_norm, Cl_norm](15)

[0142] Among them, BC represents the biochemical routine feature vector, ALT_norm represents alanine aminotransferase, AST_norm represents aspartate aminotransferase, TBIL_norm represents total bilirubin, Creatinine_norm represents serum creatinine, BUN_norm represents blood urea nitrogen, UricAcid_norm represents uric acid, TotalCholesterol_norm represents total cholesterol, LDL-C_norm represents low-density lipoprotein cholesterol, HDL-C_norm represents high-density lipoprotein cholesterol, Triglycerides_norm represents triglycerides, FastingGlucose_norm represents fasting blood glucose, Na_norm represents electrolyte sodium, K_norm represents electrolyte potassium, Ca_norm represents electrolyte calcium, and Cl_norm represents electrolyte chloride.

[0143] ALT_norm = (ALT - ALT_min) / (ALT_max - ALT_min) (16)

[0144] Among them, ALT_norm represents alanine aminotransferase, ALT_max and ALT_min are the maximum and minimum values of alanine aminotransferase respectively, and ALT represents the average value of alanine aminotransferase.

[0145] Domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of physiological index data - sex hormone level variables. Domain knowledge and feature engineering include sex hormone combination feature vectors, and the sex hormone combination feature vector is a 5-dimensional vector composed of testosterone, estradiol, luteinizing hormone, follicle-stimulating hormone, and progesterone:

[0146] SH = [Testosterone_norm, E2_norm, LH_norm, FSH_norm, Progesterone_norm](17)

[0147] Among them, SH represents the hormone combination feature vector, Testosterone_norm represents testosterone, E2_norm represents estradiol, LH_norm represents luteinizing hormone, FSH_norm represents follicle-stimulating hormone, and Progesterone_norm represents progesterone.

[0148] Domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of the output variables of the intelligent comprehensive intervention model for depressive disorders. Domain knowledge and feature engineering include intervention warning feature vectors, and the intervention warning feature vector includes safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management:

[0149] IW = [SRM_Tip, SRM_Warning, SRM_Danger, MSM_Tip, MSM_Warning, MSM_Danger, MAM_Tip, MAM_Warning, MAM_Danger, HEM_Tip, HEM_Warning, HEM_Danger, LSM_Tip, LSM_Warning, LSM_Danger](18)

[0150] Among them, IW represents the intervention warning feature, and SRM_Tip, SRM_Warning, and SRM_Danger represent the tip, warning, and danger levels of safety risk management respectively;

[0151] MSM_Tip, MSM_Warning, and MSM_Danger represent the tip, warning, and danger levels of drug safety management respectively; MAM_Tip, MAM_Warning, and MAM_Danger represent the tip, warning, and danger levels of medication compliance management respectively; HEM_Tip, HEM_Warning, and HEM_Danger represent the tip, warning, and danger levels of health education management respectively.

[0152] Please refer to Figure 3, the present invention relates to an intelligent comprehensive intervention management system for depression outside the hospital, including a clinical data collection module, a patient-side data collection module, a feature processing module, a comprehensive intervention early warning model, a comprehensive intervention management module, an intervention task management module, a medical staff-side module, a patient-side module, and a regular evaluation module. Among them, the clinical data collection module is used to collect the clinical data of depression patients in the hospital, and the in-hospital clinical data includes test and examination data and clinical symptom data; the patient-side data collection module is used to collect out-of-hospital monitoring data, and the out-of-hospital monitoring data includes physiological index monitoring data, daily behavior data, medication data, psychological training, and popular science education data of the patient outside the hospital; the feature processing module is used to integrate the in-hospital clinical data and the out-of-hospital monitoring data, and based on domain knowledge and feature engineering, process the in-hospital clinical data and the out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model; the comprehensive intervention early warning model is used to train the comprehensive intervention early warning model based on a multi-layer perceptron neural network and output early warning management level data including multiple dimensions such as safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; the comprehensive intervention management module is used to define the management content and specific management measures of comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; the intervention task management module is used to define the management content and specific management measures of comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; the medical staff-side module is used to receive specific intervention tasks and manage the patient side; at the same time, adjust the personal information, medical history information, and mental health assessment information in the clinical record based on the changes in the condition of the depression patient and the situation of follow-up visits; and evaluate the management situation, and adjust the relevant parameters of the comprehensive intervention early warning model based on the evaluation results; the patient-side module is used to receive comprehensive intervention management tasks, record daily life data, and interact with the medical staff-side; the regular evaluation module is used to evaluate the intervention situation of depression patients from 5 dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjust the risk level and intervention plan as the basis for subsequent training and management.

[0153] The comprehensive intervention management model includes an input layer, a first hidden layer, a second hidden layer, and an output layer. Among them, the input layer, which includes 208 neurons, receives a 208-dimensional input feature vector, X = [P, H, L, U, PH, M, MH, B, BC, TH, SH]; the first hidden layer, connected to the input layer, includes 128 neurons, H1 = ReLU(W1*X + b1); the second hidden layer, connected to the first hidden layer, includes 100 neurons, H2 = ReLU(W2*H1 + b2); the output layer, connected to the second hidden layer, includes 15 neurons, corresponding to different intervention types and intervention levels respectively, Y = Sigmoid(W3*H2 + b3); where X represents the input layer, H1 represents the first hidden layer, H2 represents the second hidden layer, Y represents the output layer, P represents the personal basic information feature vector, H represents the personal medical history and family medical history feature vector, L represents the lifestyle feature vector, U represents the medication data feature vector, PH represents the physical examination feature vector, M represents the mental status examination feature vector, MH represents the mental health feature vector, B represents the biochemical routine feature vector, BC represents the biochemical routine feature vector, TH represents the thyroid function feature vector, SH represents the sex hormone combination feature vector; W1, W2, and W3 are the weight matrices of the first hidden layer, the second hidden layer, and the output layer respectively, the value range of W1 is [-0.01, 0.01], the value range of W2 is [-0.15, 0.15], the value range of W3 is [-0.02, 0.02], b1, b2, and b3 are the bias vectors of the first hidden layer, the second hidden layer, and the output layer respectively, and the value range is [0, 0.01]; ReLU is the activation function, and Sigmoid is the activation function of the output layer.

[0154] The clinical data collection module collects the in-hospital clinical data of depression patients. The out-of-hospital monitoring data mainly includes collecting personal basic information, personal history and family history, medication data, physical examination data, mental status examination data, mental health status assessment data, blood routine data, biochemical routine data, thyroid function data, sex hormone level data, etc. from the in-hospital HIS (Hospital Information System), electronic medical records, examination and inspection systems, etc., and processes the data into a data format for subsequent feature engineering and model training.

[0155] The patient-side data collection module collects out-of-hospital monitoring data. The out-of-hospital monitoring data mainly includes collecting personal lifestyle data from the patient-side mobile application and wearable devices, and processes it into a data format for subsequent feature engineering and model training.

[0156] The feature processing module integrates the data from the clinical data collection module and the patient - side data collection module, and processes the data into input variables for the comprehensive intervention warning model based on MLP (Multi - layer perceptron) through domain knowledge and feature engineering.

[0157] The comprehensive intervention warning model is trained based on the MLP (Multi - layer perceptron) neural network, and outputs warning management level data including multiple dimensions such as safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management.

[0158] The comprehensive intervention management module defines the management content and specific management measures of the comprehensive intervention based on the different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management. For example, if the drug safety level is identified as a warning, information will be sent to the doctor to prompt whether the patient needs to adjust or stop taking the medicine. If the suicide risk management level is dangerous, the first - level psychological crisis intervention measures will be initiated, and the attending doctor, patient's family members, and suicide crisis intervention team will be contacted immediately.

[0159] The intervention task management module forms system - automated execution tasks based on the intervention content and management measures of the comprehensive intervention management module and sends them to the medical staff side or the patient side. For example, if the comprehensive intervention warning model outputs that the patient needs to strengthen the learning of certain aspects of health education knowledge, according to the corresponding content in the comprehensive intervention management module, the intervention task management module will generate the corresponding content and send it to the patient side. If the comprehensive intervention warning model identifies that the patient needs to strengthen certain aspects of lifestyle management, the intervention task management module will send the corresponding management tasks and tips to both the doctor side and the patient side.

[0160] The medical staff side module receives the specific intervention tasks sent by the intervention task management module and manages the patient side; at the same time, the medical staff side module can adjust the personal information, medical history information, mental health assessment information, etc. in the clinical record based on the changes in the patient's condition and the follow - up situation; at the same time, the medical staff side will also evaluate the management situation and adjust the relevant parameters of the comprehensive intervention warning model based on the evaluation results.

[0161] The patient side module receives the comprehensive intervention management tasks sent by the intervention task management module, such as reminders for medication compliance, mental health knowledge and mental training content, collects data such as exercise and sleep obtained from wearable devices, records daily life data such as diet and mood, etc., and can interact with the doctor side.

[0162] The regular evaluation module evaluates the intervention status of depression patients from five dimensions: safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjusts the risk level and intervention plan, which serves as the basis for subsequent training and management.

[0163] The intelligent comprehensive intervention management method and system for depression outside the hospital provided in this embodiment, compared with the prior art, respectively collect the in-hospital clinical data of depression patients in the hospital and the out-of-hospital monitoring data outside the hospital, and process the in-hospital clinical data and out-of-hospital monitoring data into a data format for subsequent feature engineering and model training; among them, the in-hospital clinical data includes inspection and examination data and clinical symptom data; the out-of-hospital monitoring data includes the physiological index monitoring data, daily behavior data, medication data, psychological training and science popularization education data of the patient outside the hospital; integrate the in-hospital clinical data and out-of-hospital monitoring data, and based on domain knowledge and feature engineering, process the in-hospital clinical data and out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model; train the comprehensive intervention early warning model based on the multi-layer perceptron neural network, and output early warning management level data including multiple dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; define the management content and specific management measures of the comprehensive intervention based on the different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; form a system automatic execution task based on the management content and specific management measures of the comprehensive intervention, and send the specific intervention task and comprehensive intervention management task to the medical staff side or the patient side; the medical staff side receives the specific intervention task and manages the patient side; at the same time, adjust the personal information, medical history information, and mental health assessment information in the clinical record based on the changes in the condition of the depression patient and the follow-up situation; and evaluate the management situation, and adjust the relevant parameters of the comprehensive intervention early warning model based on the evaluation results; the patient side receives the comprehensive intervention management task, records daily life data, and interacts with the medical staff side; evaluate the intervention situation of depression patients from 5 dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjust the risk level and intervention plan as the basis for subsequent training and management. The intelligent comprehensive intervention management method and system for depression outside the hospital provided in this embodiment provide a comprehensive and personalized new treatment and monitoring plan for depression patients by integrating in-hospital and out-of-hospital data and applying an advanced neural network model; use the neural network model to process the collected data, analyze the treatment response, recurrence risk, and drug safety of patients, and formulate a personalized out-of-hospital treatment and monitoring plan according to the model output, including medication guidance, improvement of living habits, mental health education, etc.; the system can monitor data changes in real time, and give early warnings and interventions for high-risk situations such as suicide tendency and medication risk; provide remote consultation and emergency contact functions to ensure that patients can get professional help in time when encountering high-risk situations; not only improve the effectiveness and safety of depression treatment, but also provide more convenient and timely support for patients.

[0164] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. An intelligent comprehensive intervention management method for depression outside the hospital, characterized in that It includes the following steps: Collect the in-hospital clinical data and out-of-hospital monitoring data of depression patients respectively in the hospital, and process the in-hospital clinical data and the out-of-hospital monitoring data into a data format for subsequent feature engineering and model training; wherein, the in-hospital clinical data includes test and examination data and clinical symptom data; the out-of-hospital monitoring data includes physiological index monitoring data, daily behavior data, medication data, psychological training and popular science education data of patients outside the hospital; Integrate the in-hospital clinical data and the out-of-hospital monitoring data, and process the in-hospital clinical data and the out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering; Train the comprehensive intervention early warning model based on a multi-layer perceptron neural network, and output early warning management level data including multiple dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; Define the management content and specific management measures of comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; Form a system automated execution task based on the management content and specific management measures of the comprehensive intervention, and send specific intervention tasks and comprehensive intervention management tasks to the medical staff side or the patient side; The medical staff side receives the specific intervention tasks and manages the patient side; at the same time, adjust the personal information, medical history information, and mental health assessment information in the clinical record based on the changes in the condition of depression patients and the follow-up situation; and evaluate the management situation, and adjust the relevant parameters of the comprehensive intervention early warning model based on the evaluation results; The patient side receives the comprehensive intervention management tasks, records daily life data, and interacts with the medical staff side; Evaluate the intervention situation of depression patients from five dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjust the risk level and intervention plan as the basis for subsequent training and management; The comprehensive intervention management model includes an input layer, a first hidden layer, a second hidden layer, and an output layer, The input layer, including 208 neurons, receives a 208-dimensional input feature vector, X = [P, H, L, U, PH, M, MH, B, BC, TH, SH]; The first hidden layer, connected to the input layer, includes 128 neurons, H1 = ReLU(W1 * X + b1); The second hidden layer, connected to the first hidden layer, includes 100 neurons, H2 = ReLU(W2 * H1 + b2); The output layer, connected to the second hidden layer, includes 15 neurons, corresponding to different intervention types and intervention levels respectively, Y = Sigmoid(W3 * H2 + b3); Among them, X represents the input layer, H1 represents the first hidden layer, H2 represents the second hidden layer, Y represents the output layer, P represents the personal basic information feature vector, H represents the personal medical history and family medical history feature vector, L represents the lifestyle feature vector, U represents the medication data feature vector, PH represents the physical examination feature vector, M represents the mental status examination feature vector, MH represents the mental health feature vector, B represents the biochemical routine feature vector, BC represents the biochemical routine feature vector, TH represents the thyroid function feature vector, SH represents the sex hormone combination feature vector; W1, W2, and W3 are the weight matrices of the first hidden layer, the second hidden layer, and the output layer respectively. The value range of W1 is [-0.01, 0.01], the value range of W2 is [-0.15, 0.15], and the value range of W3 is [-0.02, 0.02]. b1, b2, and b3 are the bias vectors of the first hidden layer, the second hidden layer, and the output layer respectively, and the value range is [0, 0.01]; ReLU is the activation function, and Sigmoid is the activation function of the output layer.

2. The intelligent comprehensive intervention management method for depression outside the hospital according to claim 1, wherein, In the step of integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the comprehensive intervention early warning model is a multi-task deep learning model based on a multi-layer perceptron neural network, with seven major data features including personal basic information variables, personal medical history and family medical history variables, lifestyle variables, medication data variables, clinical observation and evaluation data variables, mental health status data variables, and physiological index data variables as input variables, and five dimensions including safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management as output variables.

3. The intelligent comprehensive intervention management method for depression outside the hospital according to claim 2, wherein In the step of integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of personal basic information variables. The domain knowledge and feature engineering include personal basic information features. The personal basic information feature vector is a 14-dimensional vector composed of features related to age, gender, economic status, living environment, and education level: P = [A_norm, G_male, G_female, G_other, E_low, E_medium, E_high, L_urban, L_suburban, L_rura, Ed_uneducated, Ed_primary, Ed_secondary, Ed_university] Among them, P represents the personal basic information feature vector; A_norm is the value after normalizing the patient's age; G_male, G_female, and G_other represent the categorical variables of male, female, and others respectively; E_low, E_medium, and E_high represent the sample economic level data, which are divided into three categories: low income, medium income, and high income; L_urban, L_suburban, and L_rural are the living environment data, representing urban, suburban, and rural respectively; Ed_uneducated, Ed_primary, Ed_secondary, and Ed_university are the education level variables, corresponding to uneducated, primary school, secondary school, and university respectively. A_norm = (A - A_min) / (A_max - A_min) Among them, A_norm is the value after normalizing the patient's age; A_min and A_max represent the minimum and maximum values of the patient's age respectively; A represents the average value of the patient's age. The domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of personal medical history and family medical history. The domain knowledge and feature engineering include the personal medical history and family medical history feature vector. The personal medical history and family medical history feature vector is a 12-dimensional vector composed of features related to the history of past chronic diseases, the history of past infectious diseases, the number of family members with depression, and the number of family members with mental illness: H = [D_diabetes, D_hypertension, D_hear, D_copd, D_cancer, I_flu, I_tb, I_hepatitis, I_syphilis, I_hiv, F_depression_norm, F_mental_norm] Among them, H represents the personal medical history and family medical history feature vector. D_diabetes, D_hypertension, D_heart, D_copd, and D_cancer represent the common types of past chronic diseases such as diabetes, hypertension, heart disease, chronic obstructive pulmonary disease, and cancer after one-hot encoding respectively; I_flu, I_tb, I_hepatitis, I_syphilis, and I_hiv represent the common types of past infectious diseases such as influenza, tuberculosis, hepatitis, syphilis, and AIDS after one-hot encoding respectively; F_depression_norm and F_mental_norm represent the number of family members with depression and the number of family members with mental illness after normalization respectively. The domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of lifestyle variables. The domain knowledge and feature engineering include the lifestyle feature vector. The lifestyle feature vector is a 13-dimensional vector composed of features related to eating habits, smoking habits, drinking habits, exercise habits, and sleep conditions: L = [D_normal, D_meat, D_veg, D_light, D_heavy, S, R_light, R_heavy, R_none, C_norm, T_norm, D_s_norm, N_norm] Among them, L represents the lifestyle feature vector. D_normal, D_meat, D_veg, D_light, and D_heavy are respectively the classification information of eating habits; S is the classification information of smoking habits; R_light, R_heavy, and R_none are respectively the classification information of drinking habits; C is the daily exercise step information after normalization; T_norm, D_s_norm, and N_norm respectively represent the sleep-related feature information after normalization.

4. The method for intelligent comprehensive intervention management of depression outside the hospital according to claim 3, characterized in that In the step of integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing them into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of commonly used drugs for depression patients. The domain knowledge and feature engineering include a medication data feature vector, and the medication data feature vector is a 24-dimensional vector composed of features related to drug names, drug doses, days of taking, and frequencies of taking: U = [M1, M2, M3, M4, M5, M6, D_{M1}, D_{M2}, D_{M3}, D_{M4}, D_{M5}, D_{M6}, T_{M1}, T_{M2}, T_{M3}, T_{M4}, T_{M5}, T_{M6}, F_{M1}, F_{M2}, F_{M3}, F_{M4}, F_{M5}, F_{M6}] Among them, U represents the medication data feature vector. M1, M2, M3, M4, M5, and M6 respectively represent 6 drugs; D_{M1}, D_{M2}, D_{M3}, D_{M4}, D_{M5}, and D_{M6} respectively represent the doses corresponding to each drug; T_{M1}, T_{M2}, T_{M3}, T_{M4}, T_{M5}, and T_{M6} respectively represent the days of taking corresponding to each drug, and F_{M1}, F_{M2}, F_{M3}, F_{M4}, F_{M5}, and F_{M6} respectively represent the frequencies of taking corresponding to each drug; The domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of physical examination variables. The domain knowledge and feature engineering include a physical examination feature vector, and the physical examination feature vector is a 30-dimensional vector composed of features related to head and face signs, cardiovascular signs, chest and abdomen signs, and nervous system signs: PH = [FS_1, FS_2,..., FS_8, CVS_1, CVS_2,..., CVS_5, TAS_1, TAS_2,.., TAS_5, NS_1, NS_2,..., NS_12] Among them, PH represents the physical examination feature vector, and FS_1, FS_2,..., FS_8 respectively represent the head and facial signs, including premature aging, mydriasis, miosis, nystagmus, abnormal eye movement, abnormal light reflex, abnormal convergence reflex, and thyroid enlargement; CVS_1, CVS_2,..., CVS_5 respectively represent the cardiovascular signs, including elevated blood pressure, decreased blood pressure, tachycardia, bradycardia, and abnormal heart sounds; TAS_1, TAS_2,..., TAS_5 represent the chest and abdominal signs, including respiratory distress, spider nevus, hepatomegaly, splenomegaly, and ascites; NS_1, NS_2,..., NS_12 represent the nervous system signs, including low muscle strength, high muscle strength, hypomyotonia, hypermyotonia, hyperactive tendon reflex, pathological reflex, meningeal irritation sign, weakened or absent superficial reflex, weakened or absent deep reflex, sensory disturbance, hemiplegia, and paraplegia.

5. The intelligent comprehensive intervention management method for depression outside the hospital according to claim 4, wherein, In the step of integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing the in-hospital clinical data and the out-of-hospital monitoring data into the input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of the mental status examination variables. The domain knowledge and feature engineering include the mental status examination feature vector, and the mental status examination feature vector is a 71-dimensional vector composed of features related to perceptual disorders, formal thought disorders, thought content disorders, thought logic disorders, attention disorders, memory disorders, intellectual disorders, disorientation, emotional activities, volitional disorders, and lack of insight: M = [PD_1, PD_2,..., PD_6, TPD_1, TPD_2,..., TPD_13, TCD_1, TCD_2,..., TCD_11, LTD_1, LTD_2,..., LTD_6, AD_1, AD_2,..., AD_6, MD_1, MD_2,..., MD_6, ID_1,.., ID_4, OD_1, OD_2, OD_3, EA_1, EA_2,..., EA_9, VD_1, VD_2,..., VD_5, LI_1, LI_2] Among them, M represents the mental status examination feature vector, and PD_1, PD_2, ..., PD_6 respectively represent perceptual disorders, including none, auditory hallucination, visual hallucination, olfactory hallucination, gustatory hallucination, and tactile hallucination; TPD_1, TPD_2, ..., TPD_13 respectively represent formal thought disorders, including flight of ideas, thought retardation, poverty of thought, loose association, fragmented thought, incoherence of thought, thought block, thought insertion, thought pressure, circumstantiality, pathological symbolic thinking, neologism, and paralogism; TCD_1, TCD_2, ..., TCD_11 respectively represent disorders of thought content, including delusion of persecution, delusion of reference, delusion of being controlled, delusion of physical influence, delusion of grandeur, delusion of guilt, delusion of jealousy, delusion of love, delusion of non-blood relationship, thought echo, and feeling of being understood; LTD_1, LTD_2, ..., LTD_6 respectively represent disorders of thought logic, including pathological symbolic thinking, neologism, symbolic thinking, illogical thinking, paralogism, and sophistry; AD_1, AD_2, ..., AD_6 respectively represent attention disorders, including hyperprosexia, hypoprosexia, circumstantiality of attention, obtention of attention, narrowing of attention, and fixation of attention; MD_1, MD_2, ..., MD_6 respectively represent memory disorders, including hypomnesia, hypermnesia, amnesia, confabulation, fabrication, and cryptomnesia; ID_1, ..., ID_4 respectively represent intellectual disorders, including mild, moderate, severe, and profound mental retardation; OD_1, OD_2, ..., OD_3 respectively represent disorders of orientation, including disorders of person orientation, place orientation, and time orientation; EA_1, EA_2, ..., EA_9 respectively represent emotional activities, including elation, depression, anxiety, fear, emotional flatness, irritability, emotional fragility, emotional paranoia, and ambivalence; VD_1, VD_2, ..., VD_5 respectively represent disorders of will, including hypobulia, hyperbulia, psychomotor excitement, psychomotor inhibition, and strange actions or behaviors; LI_1 and LI_2 respectively represent the classification of lack of insight, 1 represents lack, and 0 represents no lack.

6. The intelligent comprehensive intervention management method for depression outside the hospital according to claim 5, characterized in that In the step of integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing the in-hospital clinical data and the out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the feature parameters, value ranges, and data processing methods of mental health status variables. The domain knowledge and feature engineering include a mental health feature vector, and the mental health feature vector is an 8-dimensional vector composed of the Young Mania Rating Scale score and the Hamilton Depression Rating Scale score: MH = [YMRS_Normal, YMRS_Mild, YMRS_Moderate, YMRS_Severe, YMRS_VerySevere, HAMD_NoDepression, HAMD_MildToModerate, HAMD_Severe] Among them, MH represents the mental health feature vector. YMRS_Normal, YMRS_Mild, YMRS_Moderate, YMRS_Severe, and YMRS_VerySevere represent the Young Mania Rating Scale scores, representing normal, mild mania, moderate mania, severe mania, and very severe mania respectively; HAMD_NoDepression, HAMD_MildToModerate, and HAMD_Severe represent the Hamilton Depression Rating Scale scores, representing no depression, mild to moderate depression, and severe depression respectively.

7. The intelligent comprehensive intervention management method for depression outside the hospital according to claim 6, characterized in that, In the step of integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing the in-hospital clinical data and the out-of-hospital monitoring data into the input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of the physiological index data - blood routine variables. The domain knowledge and feature engineering include a blood routine feature vector, and the blood routine feature vector is a 9-dimensional vector composed of red blood cell count, hemoglobin, white blood cell count, platelet count, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin content, mean corpuscular hemoglobin concentration, and red blood cell distribution width: B = [RBC_norm, Hb_norm, WBC_norm, Platelets_norm, HCT_norm, MCV_norm, MCH_norm, MCHC_norm, RDW_norm] Among them, B represents the blood routine feature vector, RBC_norm represents the red blood cell count, Hb_norm represents the hemoglobin count, WBC_norm represents the white blood cell count, Platelets_norm represents the platelet count, HCT_norm represents the hematocrit, MCV_norm represents the mean corpuscular volume, MCH_norm represents the mean corpuscular hemoglobin content, MCHC_norm represents the mean corpuscular hemoglobin concentration, and RDW_norm represents the red blood cell distribution width; RBC_norm = (RBC - RBC_min) / (RBC_max - RBC_min) Among them, RBC_norm represents the red blood cell count, RBC_max and RBC_min are the maximum and minimum values of the red blood cell count respectively, and RBC represents the average value of the red blood cell count.

8. The intelligent comprehensive off-hospital intervention management method for depression according to claim 7, characterized in that In the step of integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing the in-hospital clinical data and the out-of-hospital monitoring data into input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the characteristic parameters, value ranges and data processing methods of physiological index data-biochemical routine variables. The domain knowledge and feature engineering include a biochemical routine feature vector, and the biochemical routine feature vector is a 15-dimensional vector composed of alanine aminotransferase, aspartate aminotransferase, total bilirubin, serum creatinine, urea nitrogen, uric acid, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, triglyceride, fasting blood glucose, electrolyte sodium, electrolyte potassium, electrolyte calcium and electrolyte chloride: BC = [ALT_norm, AST_norm, TBIL_norm, Creatinine_norm, BUN_norm, UricAcid_norm, TotalCholesterol_norm, LDL-C_norm, HDL-C_norm, Triglycerides_norm, FastingGlucose_norm, Na_norm, K_norm, Ca_norm, Cl_norm] where BC represents the biochemical routine feature vector, ALT_norm represents alanine aminotransferase, AST_norm represents aspartate aminotransferase, TBIL_norm represents total bilirubin, Creatinine_norm represents serum creatinine, BUN_norm represents urea nitrogen, UricAcid_norm represents uric acid, TotalCholesterol_norm represents total cholesterol, LDL-C_norm represents low-density lipoprotein cholesterol, HDL-C_norm represents high-density lipoprotein cholesterol, Triglycerides_norm represents triglyceride, FastingGlucose_norm represents fasting blood glucose, Na_norm represents electrolyte sodium, K_norm represents electrolyte potassium, Ca_norm represents electrolyte calcium, and Cl_norm represents electrolyte chloride; ALT_norm = (ALT - ALT_min) / (ALT_max - ALT_min) where ALT_norm represents alanine aminotransferase, ALT_max and ALT_min are the maximum and minimum values of alanine aminotransferase respectively, and ALT represents the average value of alanine aminotransferase; The domain knowledge and feature engineering define the characteristic parameters, value ranges and data processing methods of physiological index data-sex hormone level variables. The domain knowledge and feature engineering include a sex hormone combination feature vector, and the sex hormone combination feature vector is a 5-dimensional vector composed of testosterone, estradiol, luteinizing hormone, follicle-stimulating hormone and progesterone: SH = [Testosterone_norm, E2_norm, LH_norm, FSH_norm, Progesterone_norm] Among them, SH represents the hormone combination feature vector, Testosterone_norm represents testosterone, E2_norm represents estradiol, LH_norm represents luteinizing hormone, FSH_norm represents follicle-stimulating hormone, and Progesterone_norm represents progesterone.

9. The intelligent comprehensive intervention management method for depression outside the hospital according to claim 8, wherein In the step of integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing the in-hospital clinical data and the out-of-hospital monitoring data into the input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering, the domain knowledge and feature engineering define the characteristic parameters, value ranges, and data processing methods of the output variables of the intelligent comprehensive intervention model for depressive disorders. The domain knowledge and feature engineering include the intervention early warning feature vector, and the intervention early warning feature vector includes safety risk management, medication safety management, medication compliance management, health education management, and lifestyle management: IW = [SRM_Tip, SRM_Warning, SRM_Danger, MSM_Tip, MSM_Warning, MSM_Danger, MAM_Tip, MAM_Warning, MAM_Danger, HEM_Tip, HEM_Warning, HEM_Danger, LSM_Tip, LSM_Warning, LSM_Danger] Among them, IW represents the intervention early warning feature, and SRM_Tip, SRM_Warning, and SRM_Danger respectively represent the tip, warning, and danger levels of safety risk management; MSM_Tip, MSM_Warning, and MSM_Danger respectively represent the tip, warning, and danger levels of medication safety management; MAM_Tip, MAM_Warning, and MAM_Danger respectively represent the tip, warning, and danger levels of medication compliance management; HEM_Tip, HEM_Warning, and HEM_Danger respectively represent the tip, warning, and danger levels of health education management.

10. An intelligent comprehensive intervention management system for depression outside the hospital, characterized in that, Including: A clinical data collection module for collecting the in-hospital clinical data of patients with depression. The in-hospital clinical data includes test and examination data and clinical symptom data; A patient-side data collection module for collecting out-of-hospital monitoring data. The out-of-hospital monitoring data includes the physiological index monitoring data, daily behavior data, medication data, psychological training, and science popularization and education data of the patient outside the hospital; A feature processing module for integrating the in-hospital clinical data and the out-of-hospital monitoring data and processing the in-hospital clinical data and the out-of-hospital monitoring data into the input variables of the comprehensive intervention early warning model based on domain knowledge and feature engineering; A comprehensive intervention early warning model is used to train the comprehensive intervention early warning model based on a multi-layer perceptron neural network and output early warning management level data including multiple dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; A comprehensive intervention management module is used to define the management content and specific management measures of comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; An intervention task management module is used to define the management content and specific management measures of comprehensive intervention based on different levels of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management; A medical staff terminal module is used to receive the specific intervention tasks and manage the patient terminal; at the same time, adjust the personal information, medical history information, and mental health assessment information in the clinical record based on the changes in the condition of the depression patients and the situation of follow-up; and evaluate the management situation, and adjust the relevant parameters of the comprehensive intervention early warning model based on the evaluation results; A patient terminal module is used to receive the comprehensive intervention management tasks, record daily life data, and interact with the medical staff terminal; A regular evaluation module is used to evaluate the intervention situation of depression patients from 5 dimensions of safety risk management, drug safety management, medication compliance management, health education management, and lifestyle management, and adjust the risk level and intervention plan as the basis for subsequent training and management; The comprehensive intervention management model includes an input layer, a first hidden layer, a second hidden layer, and an output layer, The input layer includes 208 neurons and receives a 208-dimensional input feature vector, X = [P, H, L, U, PH, M, MH, B, BC, TH, SH]; The first hidden layer is connected to the input layer and includes 128 neurons, H1 = ReLU(W1*X + b1); The second hidden layer is connected to the first hidden layer and includes 100 neurons, H2 = ReLU(W2*H1 + b2); The output layer is connected to the second hidden layer and includes 15 neurons, corresponding to different intervention types and intervention levels respectively, Y = Sigmoid(W3*H2 + b3); Among them, X represents the input layer, H1 represents the first hidden layer, H2 represents the second hidden layer, Y represents the output layer, P represents the personal basic information feature vector, H represents the personal medical history and family medical history feature vector, L represents the lifestyle feature vector, U represents the medication data feature vector, PH represents the physical examination feature vector, M represents the mental status examination feature vector, MH represents the mental health feature vector, B represents the biochemical routine feature vector, BC represents the biochemical routine feature vector, TH represents the thyroid function feature vector, SH represents the sex hormone combination feature vector; W1, W2, and W3 are the weight matrices of the first hidden layer, the second hidden layer, and the output layer respectively. The value range of W1 is [-0.01, 0.01], the value range of W2 is [-0.15, 0.15], and the value range of W3 is [-0.02, 0.02]. b1, b2, and b3 are the bias vectors of the first hidden layer, the second hidden layer, and the output layer respectively, and the value range is [0, 0.01]; ReLU is the activation function, and Sigmoid is the activation function of the output layer.

Images