Group of phenotypic variables related to mental disability level, prediction model and construction method of phenotypic variables and prediction model
By constructing a prediction model based on LASSO and random forest models, the characteristics related to mental disability levels were screened out, and the subjectivity problem in forensic mental disability level assessment was solved, and the rapid and accurate mental disability level prediction was achieved, and the litigation cost was reduced.
Patent Information
- Application Number
- CN202410160902.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-05
- Publication Date
- 2025-08-05
AI Technical Summary
The existing forensic mental disability rating standards rely on experience and subjective judgment, resulting in inconsistent identification results and increasing litigation costs. The existing TBI prognosis model is not applicable to forensic mental disability rating assessment.
The LASSO algorithm and random forest model are used to screen out the characteristics related to the mental disability level from multiple features, and a predictive model is constructed, including the cause of injury, traffic accidents, consciousness impairment, degree of self-knowledge damage, etc., and select appropriate parameters through cross-validation to build LASSO and random forest models.
It achieves rapid and accurate prediction of mental disability levels, and has high results stability and efficiency. It is suitable for different types of TBI patients, reducing identification disputes and reducing litigation costs.
Smart Images
Figure CN120432137A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of forensic medicine, and in particular to a group of phenotypic variables related to mental disability levels and a prediction model and a construction method thereof. Background Art
[0002] Traumatic brain injury (TBI) is a type of brain injury caused by a violent blow to the head that disrupts the brain's normal structure. Severe trauma not only damages the brain's structure but also disrupts its function, leading to neuropsychiatric symptoms characterized by cognitive, emotional, or behavioral abnormalities. These symptoms can be long-lasting, and severe, can lead to lifelong mental disorders that can impact an individual's life and social functioning, causing irreversible mental disability, significantly affecting their quality of life, and placing a heavy economic burden on society and families. Globally, over 50 million people suffer from TBI annually.
[0003] Mental disability refers to irreversible mental damage, resulting in a lifelong impact on an individual's life and social functioning. Mental disability assessment provides a basis for civil compensation litigation and is crucial to the economic interests of both parties. Currently, qualified forensic psychiatric experts, based on the "Classification of Disability Caused by Human Injury," categorize mental disability levels from Level 1 to Level 10 based on the severity of the individual's mental disorder and the degree of impairment in their social functioning. The interpretation of specific items in this standard is relatively general and relies heavily on the experience and subjective judgment of forensic psychiatrists. Furthermore, influenced by the desire for compensation, individuals often exaggerate and feign illness during assessments, resulting in a high rate of reassessments in forensic mental disability assessments, increased litigation costs, and compromised judicial fairness. Therefore, developing a model for predicting mental disability levels with high predictive power and ease of application is one way to address the current challenges.
[0004] Machine learning algorithms can be used to construct predictive models. Previous studies on patients with traumatic brain injury have constructed predictive models for prognostic indicators such as the Glasgow Outcome Scale (GOS). Although both the GOS and the mental disability rating are assessed some time after injury, evaluating recovery after clinical treatment and rehabilitation, there are fundamental differences between the two. The GOS is commonly used in clinical assessments of patient prognosis and is divided into five levels: death, vegetative state, severe disability, moderate disability, and good recovery. It rarely involves mental status and, in practical applications, is often divided into only two levels: good prognosis and poor prognosis. The forensic mental disability rating assesses the severity of mental disorders and the degree of impairment in social functioning, and divides disability into ten levels. Therefore, the existing GOS predictive model for TBI prognosis is not suitable for forensic mental disability rating assessment. Summary of the Invention
[0005] The purpose of the present invention is to address the above technical problems, screen for characteristics that affect the level of mental disability in TBI patients, and provide a set of phenotypic variables related to the level of mental disability and a more accurate model for predicting the level of mental disability.
[0006] In order to achieve the above object, the present invention adopts the following technical solutions:
[0007] This study uses the LASSO (Least Absolute Shrinkage and Selection Operator) algorithm to identify a set of features that significantly contribute to the mental disability rating of TBI patients from over 200 characteristics used in forensic mental disability assessments. This study also provides a set of phenotypic variables associated with mental disability rating, including: cause of injury: traffic accident, impaired consciousness, degree of impaired insight, impaired daily living abilities, psychotic symptoms, intellectual impairment, affective disturbance, disorientation, speech impairment, impaired remote memory, impaired social functioning, number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impaired recent memory.
[0008] The present invention provides a LASSO model for predicting the degree of mental disability, which is constructed based on the following set of characteristics: the cause of injury is a traffic accident, impaired consciousness, the degree of impaired self-awareness, impaired daily living ability, psychotic symptoms, intellectual impairment, emotional disorder, disorientation, speech disorder, impaired remote memory, impaired social function, the number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impaired recent memory.
[0009] The present invention provides a method for constructing the LASSO model, wherein, using features associated with the mental disability level of TBI patients, appropriate parameters are selected through cross-validation, a LASSO model for predicting the mental disability level is trained in a training set, and the model effectiveness is verified in an independent test set, wherein the features associated with the mental disability level of TBI patients include: injury cause being a traffic accident, impaired consciousness, degree of impairment of self-awareness, impairment of daily living ability, psychotic symptoms, intellectual impairment, affective disorder, disorientation, speech disorder, impairment of remote memory, impairment of social function, the number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impairment of recent memory.
[0010] The present invention provides a random forest model for predicting the degree of mental disability. The random forest model is constructed based on the following set of features: the cause of injury is a traffic accident, impaired consciousness, the degree of impaired self-awareness, impaired daily living ability, psychotic symptoms, intellectual impairment, emotional disorder, disorientation, speech disorder, impaired remote memory, impaired social function, the number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impaired recent memory.
[0011] The present invention provides a method for constructing the random forest model, wherein, using features associated with the mental disability level of TBI patients, appropriate parameters are selected through cross-validation, the random forest model for predicting the mental disability level is trained in a training set, and the model effectiveness is verified in an independent test set, wherein the features associated with the mental disability level of TBI patients include: injury cause being a traffic accident, impaired consciousness, degree of impairment of self-awareness, impairment of daily living ability, psychotic symptoms, intellectual impairment, affective disorder, disorientation, speech disorder, impaired remote memory, impaired social function, the number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impaired recent memory.
[0012] Compared with the prior art, the phenotypic variables and related prediction models of the present invention can be used to predict the level of mental disability with short time consumption, high efficiency, and good accuracy and stability of the results. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 are LASSO model variables and variable coefficients.
[0014] Figure 2 are the random forest model variables and their importance. DETAILED DESCRIPTION
[0015] The technical solutions of the present invention are further described in detail below with reference to specific embodiments, but the present invention is not limited to the following embodiments.
[0016] 1. Variable Collection and Processing
[0017] 1. Collect relevant information such as the general condition, craniocerebral injury medical history, identification, psychological testing, etc. of the TBI patient and verify the information to ensure it is correct.
[0018] The collected information includes: patient's gender, age, education level, medical history related to the nervous system before injury, history of brain trauma, history of cerebrovascular disease before injury, history of mental illness before injury, history of epilepsy before injury, use of anti-epileptic drugs, drug abuse history, smoking history, history of hypertension, use of hypertension treatment drugs, blood pressure control, history of diabetes, use of diabetes treatment drugs, blood sugar control, gout, family history of mental illness, date of injury, type of injury (including but not limited to traffic accident, violent injury, fall from height, fall, etc.), location of injury, Location at the time of injury, vehicle used at the time of injury, direction of impact at the time of injury, damage to the other vehicle's vehicle type in the traffic accident, extent of violent injury suffered, impaired consciousness at the time of assessment, disorientation, willingness to cooperate during psychiatric examination, presence of exaggerated performance, attention deficit, speech disorder, slow reaction, intellectual impairment, impairment of remote memory, impairment of recent memory, degree of immediate memory loss, amnesia, psychotic symptoms, affective disorder, increased or decreased volitional activity, degree of impairment of self-awareness, stress disorder, MMSE score, Wechsler Intelligence Test, Wechsler Memory Quotient, degree of cooperation, whether wrong answers were given intentionally, drowsiness / coma after injury, vomiting , dilated pupils, emotional instability, fear, tension, excitement, delirium, amnesia, scalp hematoma, skull fracture, epidural hematoma, subdural hematoma, cerebral contusion, brain hematoma, subarachnoid hemorrhage, intraventricular hemorrhage, diffuse axonal injury, brain swelling, brain herniation, hydrocephalus, intracranial pneumocephalus, cerebrospinal fluid leakage, sinusitis, sinus effusion, white matter lesions, traumatic cerebral ischemia and cerebral infarction, physical injury, cerebral atrophy, softening foci in the brain (including but not limited to left frontal, left parietal, left temporal, left occipital, right frontal, right parietal, right temporal, right occipital, left basal ganglia, right basal ganglia, etc. and softening foci number of cerebral lobes), skull defect or repair, aphasia, limb weakness, dementia, pseudodementia, abnormal EEG, impaired daily living activities (ADL) (including but not limited to using public vehicles, walking, cooking, doing housework, taking medicine, eating, dressing, washing, shopping, going to the toilet on time, using entertainment electronic devices, using communication tools to communicate with the outside world, etc.), impaired social function, behavioral tendencies, impaired going out, intellectual impairment, social withdrawal, impaired occupation and work, impaired marital function, impaired parental function, impaired social activities outside the family, impaired family function, impaired personal self-care ability, etc.
[0019] 2. The collected information is categorized into four types: dichotomous variables, continuous variables, categorical variables with levels, and categorical variables without levels. The above information is digitized according to the definitions: the two factors in the dichotomous variable are set to "1" and "0"; the numerical code of the continuous variable is equal to its numerical value; the different levels of the categorical variable with levels are represented by numbers and analyzed as continuous variables; the categorical variables without levels are treated as dummy variables, and the categorical variables with multiple factors are constructed into multiple dummy variables that only take "0" and "1" as the values. "0" represents the failure to meet the factor, and "1" represents the compliance with the factor.
[0020] 2. Variable Screening
[0021] All collected variables were put into the model for screening.
[0022] Because random sampling is involved, the results obtained under different random seeds may be different, making the constructed LASSO model unstable. Therefore, in order to build a stable model, the present invention uses 200 seeds, each of which randomly extracts 80% of the samples from the training set each time. For each randomly extracted sample, 1000 prediction models are constructed (using 1000 seeds), and variables that appear in more than 10% of these 1000 prediction models are found. These variables are used to build a prediction model for the current sample. After constructing 200 prediction models, variables with a frequency of occurrence greater than 75% are identified based on the number of variables selected in the model and the variable coefficients. Based on these variables, 1,000 prediction models were randomly constructed using all training set samples. Fourteen variables appeared stably (occurrence ratio > 50%) (namely: injury cause was a traffic accident, impaired consciousness, degree of impaired self-awareness, impaired daily living ability, psychotic symptoms, intellectual impairment, emotional disorder, disorientation, speech disorder, impaired remote memory, impaired social function, number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impaired recent memory). These variables were the candidate variables for the final prediction model.
[0023] 3. Model Construction
[0024] 1. Construct LASSO prediction model
[0025] Based on the 14 variables identified as contributing significantly to the level of mental disability, the LASSO prediction model was constructed using the "glm" package in R. In LASSO, λ is a non-negative hyperparameter that controls the intensity of the penalty; a larger λ value results in a stronger penalty term. Ten-fold cross-validation was used to select the optimal λ value. LASSO is constructed by finding a set of variable coefficients that minimizes the following formula:
[0026]
[0027] The dependent variable is Y, the selected variable is X, β is the coefficient of X, and λ is the penalty parameter.
[0028] 2. Build a random forest prediction model
[0029] Based on the 14 variables that contributed most to the mental disability level selected above, a random forest prediction model was constructed using the “randomForest” package in R software. The parameters mtry(1,2,3,4,5,6), nodesize(2,3,4,5), nsplit(1,2,3,4,5) and ntree(200,300,400,500,600,700,800,900,1000) were used in the construction of the random forest model. Five-fold cross validation was used to find the maximum R 2 The model is constructed by combining the parameters of .
[0030] Table 1 below shows the definitions of the variables selected in the model and their coefficients in the LASSO model and their importance in the random forest model.
[0031] Table 1
[0032]
[0033] Figure 1 The LASSO model variables and their coefficients are shown.
[0034] Figure 2 The random forest model variables and their importance are shown.
[0035] IV. Model Application
[0036] By collecting a small amount of variable information required by the above-mentioned LASSO prediction model and random forest prediction model, and substituting them into the constructed LASSO prediction model and random forest prediction model respectively, the predicted mental disability level can be obtained, and the results are presented on a scale of 1 to 10. The whole process is time-saving, efficient, accurate and stable.
[0037] The present invention uses the mean absolute deviation (MAD) and the coefficient of determination (R 2 ) to evaluate the model effectiveness, MAD is used to reflect the prediction error, R 2 The LASSO prediction model constructed by the present invention was verified in an independent test set and showed that the prediction efficiency of the model was MAD = 0.607, R 2=0.836. The model's prediction accuracy in the mild TBI population (GCS 13-15 points) was higher than that in the severe TBI (GCS 3-8 points) and moderate TBI (GCS 9-12 points) populations (MAD 轻度TBI =0.436,R 2 轻度TBI =0.903; MAD 中度TBI =0.471,R 2 中度TBI =0.838; MAD 重度TBI =0.770,R 2 重度TBI =0.793); the prediction accuracy in the non-re-identification sample is higher than its accuracy in the re-identification sample (disputed cases) (MAD 非重鉴定 =0.604, R 2 非重鉴定 =0.863; MAD 重鉴定 =0.617, R 2 重鉴定 =0.747); The prediction accuracy of the model in test sample set 1 (2017-2019) is slightly higher than that in test sample set 2 (2021-2022) (MAD test1 =0.607, R 2 test1 =0.836; MAD test2 =0.715, R 2 test2 =0.818).
[0038] The random forest prediction model constructed by the present invention was verified in an independent test set, and the prediction efficiency of the model was MAD=0.581, R 2 =0.873. The model's prediction accuracy in the mild TBI (GCS 13-15) population was higher than that in the severe TBI (GCS 3-8) and moderate TBI (GCS 9-12) populations (MAD 轻度TBI =0.463,R 2 轻度TBI =0.931; MAD 中度TBI =0.516,R 2 中度TBI =0.811; MAD 重度TBI =0.681,R 2 重度TBI =0.848); the prediction accuracy in the non-re-identification sample is higher than its accuracy in the re-identification sample (disputed cases) (MAD 非重鉴定 =0.569, R 2 非重鉴定=0.899; MAD 重鉴定 =0.624, R 2 重鉴定 =0.743); The prediction accuracy of the model in test sample set 1 (2017-2019) is slightly higher than that in test sample set 2 (2021-2022) (MAD test1 =0.581, R 2 test1 =0.873; MAD test2 =0.686, R 2 test2 =0.842).
[0039] The present invention demonstrates that, in the TBI population, the following characteristics affect the prognosis and degree of mental disability in TBI patients, with TBI patients with these characteristics experiencing more severe mental disability than those without these characteristics: injury cause being a traffic accident, impaired consciousness, degree of impaired insight, impaired ability to perform daily activities, psychotic symptoms, intellectual impairment, affective disturbance, disorientation, speech impairment, impaired remote memory, impaired social functioning, the number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impaired recent memory.
[0040] In summary, the present invention has achieved for the first time the use of specific indicators to objectively evaluate the degree of mental disability, which can provide a basis for establishing a more scientific and reasonable forensic mental disability objective assessment technology, and help to gain a deeper understanding of the characteristics and phenotypes of people with craniocerebral injury.
Claims
1. A set of phenotypic variables related to the degree of mental disability, characterized by: The phenotypic variables include: the cause of injury is a traffic accident, impaired consciousness, degree of impairment of self-awareness, impairment of daily living ability, psychotic symptoms, intellectual impairment, affective disorder, disorientation, speech disorder, impairment of remote memory, impairment of social function, number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impairment of recent memory.
2. A LASSO model for predicting the level of mental disability, characterized by: The LASSO model is constructed based on the following set of characteristics: the cause of injury is a traffic accident, impaired consciousness, degree of impaired self-awareness, impaired daily living ability, psychotic symptoms, intellectual impairment, affective disorder, disorientation, speech disorder, impaired remote memory, impaired social function, number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impaired recent memory.
3. The method for constructing the LASSO model according to claim 2, wherein: The method comprises the following steps: collecting TBI patient information, using features associated with the mental disability level of TBI patients, selecting appropriate parameters through cross-validation, training a LASSO model for predicting the mental disability level in a training set, and validating the model effectiveness in an independent test set; wherein the features associated with the mental disability level of TBI patients include: injury cause being a traffic accident, impaired consciousness, degree of impairment of self-perception, impairment of daily living ability, psychotic symptoms, intellectual impairment, affective disorder, disorientation, speech disorder, impairment of remote memory, impairment of social function, the number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impairment of recent memory.
4. A random forest model for predicting mental disability level, characterized by: The random forest model was constructed based on the following set of features: cause of injury was a traffic accident, impaired consciousness, degree of impaired self-awareness, impaired daily living ability, psychotic symptoms, impaired intelligence, affective disorder, disorientation, speech disorder, impaired remote memory, impaired social function, number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impaired recent memory.
5. The method for constructing a random forest model according to claim 4, characterized in that: The method comprises the following steps: collecting TBI patient information, using features associated with the mental disability level of TBI patients, selecting appropriate parameters through cross-validation, training a random forest model for predicting the mental disability level in a training set, and validating the model effectiveness in an independent test set; wherein the features associated with the mental disability level of TBI patients include: injury cause being a traffic accident, impaired consciousness, degree of impairment of self-perception, impairment of daily living ability, psychotic symptoms, intellectual impairment, affective disorder, disorientation, speech disorder, impairment of remote memory, impairment of social function, the number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impairment of recent memory.
6. Application of a set of phenotypic variables related to mental disability level in constructing a mental disability level prediction model, characterized in that: The phenotypic variables include: the cause of injury is a traffic accident, impaired consciousness, degree of impairment of self-awareness, impairment of daily living ability, psychotic symptoms, intellectual impairment, affective disorder, disorientation, speech disorder, impairment of remote memory, impairment of social function, number of cerebral lobes with softening lesions, increased or decreased volitional activity, and impairment of recent memory.