Enhanced prediction method for infection risk of HIV / AIDS patient

Abstract

The invention discloses an enhanced prediction method for the infection risk of an HIV / AIDS patient, and relates to the technical field of infection risk prediction. According to the method, by introducing feature interaction and feature selection, redundant features are reduced, the robustness, stability and interpretability of the prediction model can be improved, the infection risk of HIV / AIDS patients can be predicted more accurately, a more reliable prediction result is provided for clinicians, making of a more effective treatment scheme and early intervention measures is facilitated, and the clinical application prospect is wide. The infection risk of HIV / AIDS patients in the treatment process is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of infection risk prediction, and in particular to an enhanced prediction method for infection risk of HIV / AIDS patients. Background Art

[0002] In the clinical management of HIV / AIDS patients, HIV / AIDS patients are at risk of contracting other opportunistic infections (such as pneumonia, tuberculosis, Pneumocystis carinii pneumonia, etc.) or diseases in a certain period of time in the future. These opportunistic infections are more likely to occur when the immune system is severely weakened by HIV, posing a major threat to the health and survival rate of patients. Therefore, infection risk prediction is crucial for early intervention and treatment optimization. Traditional statistical methods and simple machine learning models have certain limitations when dealing with high-dimensional and complex interactive features, resulting in less than ideal prediction accuracy. The existing HIV / AIDS patient infection risk prediction model has low accuracy when processing data with multiple complex features, lacks effective feature selection and feature interaction mechanisms, and makes it difficult to improve the performance of the risk prediction model. Therefore, developing a more advanced and accurate prediction model can significantly improve the effectiveness of clinical decision-making. Summary of the invention

[0003] The purpose of the present invention is to provide an enhanced prediction method for the infection risk of HIV / AIDS patients to improve the above technical problems.

[0004] In order to achieve the above-mentioned object of the invention, the embodiment of the present invention provides the following technical solutions:

[0005] An enhanced prediction method for HIV / AIDS patient infection risk, comprising:

[0006] S1. Acquire clinical data of HIV patients or AIDS patients and perform preprocessing to obtain preprocessed clinical data;

[0007] S2. Construct an enhanced infection risk prediction model;

[0008] S3. Inputting the pre-processed clinical data into the enhanced infection risk prediction model, and outputting the infection risk prediction result of the HIV or AIDS patient.

[0009] Furthermore, the clinical data of the HIV or AIDS patients include demographic data, HIV-specific data, routine physiological data and influencing factors; the demographic data include the age, gender, race and geographical location of the HIV or AIDS patients; the HIV-specific data include the CD4 cell count, disease load, medication record and efficacy of antiretroviral treatment, and previous history of opportunistic infections of the HIV or AIDS patients; the disease load includes HIV level and RNA level; the routine physiological data include body temperature, blood pressure, heart rate, routine blood data, biochemical indicators, and nutritional status of the HIV or AIDS patients; the influencing factors include smoking, drinking, drug abuse history, socioeconomic status and mental and psychological status of the HIV or AIDS patients; the routine blood data include white blood cell count, red blood cell count, platelet count; the biochemical indicators include liver function and kidney function; the nutritional status includes serum albumin level.

[0010] Furthermore, the enhanced infection risk prediction model adopts a logistic regression model.

[0011] Furthermore, the construction of the enhanced infection risk prediction model includes the following steps:

[0012] S2-1, based on the formula:

[0013]

[0014] Construct the first logistic regression model P0(Y=1|X); where Y represents the prediction result, X represents the feature data set, e represents the natural constant, β0 represents the bias term, β1, β n represents the regression coefficient, X1, X n Respectively represent the first feature data and the nth feature data;

[0015] S2-2, obtaining historical clinical data of HIV patients or AIDS patients and preprocessing them to obtain preprocessed historical clinical data;

[0016] S2-3, inputting the pre-processed historical clinical data into the first logistic regression model, processing it through a recursive feature elimination algorithm, and obtaining corresponding selected feature data;

[0017] S2-4, processing the acquired feature data by cross-validation technology and regularization method to obtain corresponding interactive feature data;

[0018] S2-5, combining the interactive feature data and the preprocessed historical clinical data to obtain combined feature data;

[0019] S2-6. Based on the merged feature data and the first logistic regression model, a second logistic regression model is constructed, and the corresponding historical infection risk prediction result is output;

[0020] S2-7. Calculate the loss function corresponding to the second logistic regression model;

[0021] S2-8. Based on the loss function, adjusting the regression coefficient of the second logistic regression model by a gradient descent method to obtain a third logistic regression model;

[0022] S2-9. Repeat S2-6 to S2-8 until the preset number of iterations is reached to complete the construction of the enhanced infection risk prediction model.

[0023] Furthermore, the pre-processed historical clinical data is input into the first logistic regression model, and processed by a recursive feature elimination algorithm to obtain corresponding selected feature data, including the following steps:

[0024] S2-3-1, initializing the first logistic regression model;

[0025] S2-3-2, inputting the pre-processed historical clinical data into the first logistic regression model to obtain corresponding regression coefficients;

[0026] S2-3-3, selecting the pre-processed historical clinical data corresponding to the maximum regression coefficient, and adding it to the feature set as feature data; the initial number of features in the feature set is 0;

[0027] S2-3-4, removing the characteristic data from the pre-processed historical clinical data to obtain updated historical clinical data;

[0028] S2-3-5, calculating the p-value of the regression coefficient corresponding to the feature data added to the feature set;

[0029] S2-3-6, determining whether the p value is less than a threshold value, if so, retaining the feature data to the feature set, and increasing the number of features in the feature set by 1; otherwise, removing the feature data from the feature set;

[0030] S2-3-7. Input the updated historical clinical data into the first logistic regression model, repeat S2-3-2 to S2-3-6 until the number of feature data in the feature set reaches a preset feature number threshold, and obtain the selected feature data.

[0031] Furthermore, the threshold in S2-3-6 is 0.05.

[0032] Furthermore, the formula of the second logistic regression model is:

[0033]

[0034] Among them, σ(·) represents the sigmoid function, n represents the total number of feature data, β i represents the regression coefficient corresponding to the i-th merged feature data, β ij represents the regression coefficient corresponding to the interaction feature corresponding to the i-th feature data and the j-th feature data, x i represents the i-th merged feature data, x j Represents the j-th merged feature data.

[0035] Furthermore, the formula of the loss function Loss is:

[0036]

[0037] Among them, y i represents the label corresponding to the i-th historical infection risk prediction result, P(y i |X i ) represents the i-th historical infection risk prediction result.

[0038] The beneficial effects of the present invention are:

[0039] The present invention introduces feature interaction and feature selection and reduces redundant features, thereby improving the robustness, stability and interpretability of the prediction model, more accurately predicting the infection risk of HIV / AIDS patients, providing clinicians with more reliable prediction results, and helping to formulate more effective treatment plans and early intervention measures, thereby reducing the infection risk of HIV / AIDS patients during treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 It is a flow chart of the invention in the embodiment of the present invention. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0043] Embodiment 1:

[0044] See also Figure 1 This embodiment provides an enhanced prediction method for HIV / AIDS patient infection risk, which includes:

[0045] S1. Acquire clinical data of HIV patients or AIDS patients and perform preprocessing to obtain preprocessed clinical data;

[0046] The clinical data of the HIV or AIDS patients include demographic data, HIV-specific data, routine physiological data and influencing factors; the demographic data include the age, gender, race and geographical location of the HIV or AIDS patients; the HIV-specific data include the CD4 cell count, disease load, medication record and efficacy of antiretroviral treatment, and previous history of opportunistic infections of the HIV or AIDS patients; the disease load includes HIV level and RNA level; the routine physiological data include the body temperature, blood pressure, heart rate, routine blood data, biochemical indicators, and nutritional status of the HIV or AIDS patients; the influencing factors include the smoking, drinking, drug abuse history, socioeconomic status and mental and psychological state of the HIV or AIDS patients; the routine blood data include white blood cell count, red blood cell count, platelet count; the biochemical indicators include liver function and kidney function; the nutritional status includes serum albumin level.

[0047] The preprocessing includes data cleaning and standardization.

[0048] S2. Construct an enhanced infection risk prediction model;

[0049] The enhanced infection risk prediction model adopts a logistic regression model.

[0050] The construction of the enhanced infection risk prediction model comprises the following steps:

[0051] S2-1, based on the formula:

[0052]

[0053] Construct the first logistic regression model P0(Y=1|X); where Y represents the prediction result, X represents the feature data set, e represents the natural constant, β0 represents the bias term, β1, β n represents the regression coefficient, X1, X n Respectively represent the first feature data and the nth feature data;

[0054] S2-2, obtaining historical clinical data of HIV patients or AIDS patients and preprocessing them to obtain preprocessed historical clinical data;

[0055] S2-3, inputting the pre-processed historical clinical data into the first logistic regression model, processing it through a recursive feature elimination algorithm, and obtaining corresponding selected feature data;

[0056] The pre-processed historical clinical data is input into the first logistic regression model, and processed by a recursive feature elimination algorithm to obtain corresponding selected feature data, comprising the following steps:

[0057] S2-3-1, initializing the first logistic regression model;

[0058] S2-3-2, inputting the pre-processed historical clinical data into the first logistic regression model to obtain corresponding regression coefficients;

[0059] S2-3-3, selecting the pre-processed historical clinical data corresponding to the maximum regression coefficient, and adding it to the feature set as feature data; the initial number of features in the feature set is 0;

[0060] S2-3-4, removing the characteristic data from the pre-processed historical clinical data to obtain updated historical clinical data;

[0061] S2-3-5. Calculate the p-value of the regression coefficient corresponding to the feature data added to the feature set. The corresponding formula is:

[0062] p=P(T>|t|);

[0063] Where t represents the statistic calculated for each feature (including interaction features), i.e., t distribution, T represents a t-distributed random variable, |·| represents the absolute value, and P(T>|t|) represents the probability that the random variable T in the t distribution takes a value greater than |t| under the null hypothesis (i.e., the feature has no actual effect in the model).

[0064] S2-3-6, determining whether the p value is less than a threshold value, if so, retaining the feature data to the feature set, and increasing the number of features in the feature set by 1; otherwise, removing the feature data from the feature set;

[0065] The threshold in S2-3-6 is 0.05.

[0066] S2-3-7. Input the updated historical clinical data into the first logistic regression model, repeat S2-3-2 to S2-3-6 until the number of feature data in the feature set reaches a preset feature number threshold, and obtain the selected feature data.

[0067] The recursive feature elimination algorithm selects feature data related to infection risk and selects feature data with significance (i.e., the p value meets the threshold) for risk prediction, which can improve the efficiency of the prediction model in risk prediction.

[0068] S2-4, processing the acquired feature data by cross-validation technology and regularization method to obtain corresponding interactive feature data;

[0069] The introduction of interactive features further expands the feature space, which means that the prediction model has more parameters that can be adjusted, which may better fit the data. Since the introduction of interactive features will also increase the complexity of the model and thus increase the risk of overfitting, cross-validation techniques and regularization methods (such as L1 and L2 regularization) are used to select appropriate feature combinations and prevent overfitting of the prediction model.

[0070] Feature combinations such as:

[0071] Age and sex: Age and sex interact to affect the risk of certain diseases, with some diseases being more common in men or women at certain ages.

[0072] Age and CD4 cell count; CD4 cell count is an important indicator of immune system function; age may affect the health of the immune system, so the interactive characteristics of age and CD4 cell count may have important implications for disease risk.

[0073] Geographic location and viral load; Geographic location may affect the chance of exposure to pathogens and the accessibility of medical resources. The viral load levels may vary in different regions. Therefore, the interactive characteristics of geographic location and viral load may affect the risk of infection.

[0074] Sex and viral load: Sex may affect how the disease develops and spreads, so the interactive characteristics of sex and viral load may help to more accurately predict the risk of infection.

[0075] Ethnicity and CD4 cell count: Different ethnic groups may have genetic differences that affect immune response, so the interactive characteristics of ethnicity and CD4 cell count may be helpful in prediction models.

[0076] S2-5, combining the interactive feature data and the preprocessed historical clinical data to obtain combined feature data;

[0077] S2-6. Based on the merged feature data and the first logistic regression model, a second logistic regression model is constructed, and the corresponding historical infection risk prediction result is output;

[0078] The formula of the second logistic regression model is:

[0079]

[0080] Among them, n represents the total number of feature data, β i represents the regression coefficient corresponding to the i-th merged feature data, β ij represents the regression coefficient corresponding to the interaction feature corresponding to the i-th feature data and the j-th feature data, x i represents the i-th merged feature data, x j represents the j-th merged feature data, σ(·) represents the sigmoid function,

[0081] The formula of the second logistic regression model introduces a quadratic term compared to the formula of the first logistic regression model. When there is a nonlinear relationship between the feature and the target variable, the logistic regression model may not fit the data well, resulting in reduced prediction performance. Therefore, the quadratic term is introduced. It can avoid problems that occur when there is a nonlinear relationship between features and the target variable.

[0082] S2-7. Calculate the loss function corresponding to the second logistic regression model; the loss function adopts a negative log-likelihood loss function.

[0083] The formula of the loss function Loss is:

[0084]

[0085] Among them, y i represents the label corresponding to the i-th historical infection risk prediction result, P(y i |X i ) represents the i-th historical infection risk prediction result.

[0086] S2-8. Based on the loss function, adjusting the regression coefficient of the second logistic regression model by a gradient descent method to obtain a third logistic regression model;

[0087] Gradient descent is an optimization algorithm used to minimize the loss function. The core idea is to gradually reduce the loss function by repeatedly adjusting the regression coefficient. The corresponding process is:

[0088] Initialize the current second logistic regression model and set the corresponding regression coefficients, for example, initialize all regression coefficients to zero;

[0089] Calculate the predicted probabilities of the current second logistic regression model;

[0090] The gradient of the loss function is calculated based on the predicted probability, and the corresponding formula is:

[0091]

[0092] in, represents the gradient of the loss function, N represents the total number of samples, which is usually used to average the sum of the loss function, P(y i |X i ) represents the predicted probability of the current second logistic regression model, x ij Represents the jth eigenvalue of the i-th sample, which is used to calculate the specific feature of the gradient.

[0093] Based on the obtained gradient, update the corresponding regression coefficient. The corresponding formula is:

[0094]

[0095] Among them, α represents the intercept, β j represents the retrospective coefficient before updating, β j : represents the updated regression coefficient.

[0096] Repeat the above steps until the loss function converges or the preset number of iterations is reached.

[0097] S2-9. Repeat S2-6 to S2-8 until the preset number of iterations is reached to complete the construction of the enhanced infection risk prediction model.

[0098] S3. Inputting the pre-processed clinical data into the enhanced infection risk prediction model, and outputting the infection risk prediction result of the HIV or AIDS patient.

[0099] Embodiment 2:

[0100] The clinical data of an infected patient were selected, with an age of 45; gender of male; ethnicity of Han nationality; geographical location of the city; CD4 cell count of 350; disease load of 10,000; ART efficacy of high efficiency; previous history of infection of yes; body temperature of 37; blood pressure of 120; heart rate of 80; white blood cell count of 6.5; red blood cell count of 4.8; platelet count of 180; liver function of normal; kidney function of normal; serum albumin of 4.2; smoking of yes; drinking of yes; drug abuse of no; economic status of medium; and mental state of normal.

[0101] The clinical data of the patient is input into the first logistic regression model (the existing logistic regression model) for risk prediction, and the corresponding formula is:

[0102] Logic(P)=-2+0.01·45+0.2·1+0.1·0+0.1·1-0.01·350+

[0103] 0.00005·10000-1.0·1+1.0·1+0.1·37+0.01·120+0.02·80

[0104] -0.1·6.5-0.1·4.8

[0105] =0.33;

[0106]

[0107] It can be seen that the corresponding value of the infection risk prediction result of this patient is 0.581,

[0108] Among them, the values ​​of the regression coefficients corresponding to the characteristic parameters of the bias term β0 are: β0 = 0, β 年龄 =0.05, β 性别 =-0.1,β 地理位置 =0.02, β 病载量 =0.001, β ART =0.2,β 感染史 =0.3,β 体温 =-0.02, β 血压 =0.01,β 心率 =0.03, β 白细胞 =-0.05, β 红细胞 =0.04, β 血小板 =0.002, β 肝功能 =0.1,β 肾功能 =-0.2, β 白蛋白 =0.5,β 吸烟 =0.6,β 饮酒 =-0.7, β 药物滥用 =0.8,β经济 =-0.5, β 精神 =0.01.

[0109] The clinical data of the patient is processed by the technical solution of the present invention, and the corresponding process is:

[0110] After being processed by the recursive feature elimination algorithm, the selected feature data is obtained, namely:

[0111] Age: 45, Gender: Male (0), Ethnicity: Han (0), Geographical location: Urban (0), CD4 cell count: 350, Disease load: 10000, ART efficacy: Highly effective (1), Previous infection history: Yes (1), Body temperature: 37, Blood pressure: 120, Heart rate: 80, White blood cell count: 6.5, Red blood cell count: 4.8, Platelet count: 180, Liver function: Normal (1), Renal function: Normal (1), Serum albumin: 4.2, Smoking: Yes (1), Drinking: No (0), Drug abuse: No (0), Economic status: Medium (1), Normal mental state (1).

[0112] The cross-validation technique and regularization method are used for interaction, and the interaction characteristics obtained are:

[0113] Age × sex = 45 × 0 = 0;

[0114] Age × CD4 cell count = 45 × 350 = 15750;

[0115] CD4 cell count × disease load = 350 × 10,000 = 3,500,000;

[0116] CD4 cell count × ART efficacy = 350 × 1 = 350;

[0117] Disease load × smoking = 10000 × 1 = 10000;

[0118] Red blood cell count × platelet count = 4.8 × 180 = 864;

[0119] Red blood cell count × white blood cell count × platelet count = 4.8 × 6.5 × 180 = 5616;

[0120] Liver function × kidney function × serum albumin = 1 × 1 × 4.2 = 4.2;

[0121] Body temperature × blood pressure × heart rate = 37 × 120 × 80 = 355200;

[0122] Age × red blood cell count × platelet count = 45 × 4.8 × 180 = 38880;

[0123] Red blood cell count × white blood cell count × platelet count × serum albumin = 4.8 × 6.5 × 180 × 4.2 = 23587.2;

[0124] Liver function × kidney function × serum albumin × mental state = 1 × 1 × 4.2 × 1 = 4.2;

[0125] Age × red blood cell count × platelet count × CD4 cell count = 45 × 4.8 × 180 × 350 = 1,360,800 and other interactive features.

[0126] The above interaction features are represented by vectors, that is, the interaction feature vector is:

[0127]

[0128] The interactive feature vector is added to the clinical data to obtain the merged feature vector, namely:

[0129] x1=(45,0,0,0,350,10000,1,1,37,120,80,6.5,4.8,180,1,1,4.2,1,0,0,1,1,

[0130] 0,45,0,15750,1665,3500000,350,10000,864,1,4.8,0,0,3500000,

[0131] 199800,5616,4.2,355200,38880,10000,0,0,0,3500000,15984000,

[0132] 23587.2,4.2,1360800,0,0,0,157500000).

[0133] The above merged feature vector is input into the enhanced infection risk prediction model. The corresponding intercept is 4, and the corresponding regression coefficient is:

[0134]

[0135] Calculate the z value, that is:

[0136] z=-4+z1+z2;

[0137] Among them, z1 is Right now:

[0138] z1=(0.05×45)+(-0.1×0)+(-0.03×0)+(0.02×0)+(0.004×350)+(0.0001× 10000)+(0.2×1)+(0.3×1)+(-0.02×37)+(0.01×120)+(0.03×80)+(-0.05 ×6.5)+(0.04×4.8)+(0.002×180)+(0.1×1)+(-0.2×1)+(0.5×4.2)+(0.6× 1)+(-0.7×0)+(0.8×0)+(-0.5×1)+(1×1)+(0.01×0)+(0.02×45)+(0.05×45 2 )+(-0.1×0 2 )+(-0.03×0 2 )+(0.02×0 2 )+(0.004×350 2 )+(0.0001×10000 2 )+(0.2×1 2 )+(0.3×1 2 )+(-0.02×37 2 )+(0.01×120 2 )+(0.03×80 2 )+(-0.05×6.5 2 )+(0.04×4.8 2 )+(0.002×180 2 )+(0.1×1 2 )+(-0.2×1 2 )+(0.5×4.2 2 )+(0.6×1 2 )+(-0.7×0 2 )+(0.8×0 2 )+(-0.5×1 2 )+(1×1 2 )+(0.01×0 2 )+(0.02×45 2 );

[0139] z2 is Right now:

[0140]

[0141] thereby:

[0142] z=-4+2.25+0+0+0+1.4+1+0.2+0.3-0.74+

[0143] 1.2+2.4-0.325+0.192+0.36+0.1-0.2+2.1+

[0144] 0.6+0+0-0.5+1+0+0.9+31.5+4.995+14000+

[0145] 1.75+60+6.048+0.008+0.0432+0+0+105000+

[0146] 7992+280.8+0.252+24864+3110.4+900+0+0+

[0147] 0+455000+2237760+3538.08+0.672+231336+

[0148] 0+0+0+33075000

[0149] =35772618.2072

[0150] The infection risk prediction results of this patient can be obtained:

[0151]

[0152] It can be seen that the infection risk prediction result of the present invention is greater than the infection risk prediction result of the existing logistic regression model, and is closer to the actual situation, which further proves the prediction accuracy of the enhanced infection risk prediction model in the present invention.

[0153] Embodiment 3:

[0154] In order to further improve the prediction accuracy of the enhanced infection risk prediction model, the enhanced infection risk prediction model in the present invention includes a support vector machine and a logistic regression model, and the corresponding steps are:

[0155] P1. Obtain clinical data of HIV patients or AIDS patients and preprocess them to obtain preprocessed clinical data;

[0156] P2. Perform feature analysis on the preprocessed clinical data to obtain linear feature data and nonlinear feature data;

[0157] The preprocessed clinical data are encoded to obtain encoded clinical data; the relationship between the encoded clinical data and the risk of infection is evaluated using the correlation coefficient method, the chi-square test method, or the mutual information method, and based on the obtained relationship, feature selection is performed to obtain linear feature data and nonlinear feature data;

[0158] Among them, linear characteristic data include age, CD4 cell count, disease load, body temperature, blood pressure, heart rate, routine blood data, and biochemical indicators of HIV or AIDS patients, and the rest are nonlinear characteristic data.

[0159] P3. Constructing a second enhanced infection risk prediction model; the second enhanced infection risk prediction model includes a support vector machine and a logistic regression model;

[0160] The construction process of the logistic regression model adopts the same method as S2.

[0161] The construction process of the support vector machine includes:

[0162] P3-1. Obtain historical nonlinear feature data and its labels, and perform preprocessing;

[0163] P3-2. Build the initial support vector machine and define the hyperparameter space;

[0164] P3-3. Input the historical nonlinear feature data and its labels into the initial support vector machine, and process the kernel function and penalty coefficient of the initial support vector machine through the grid search algorithm based on the hyperparameter space. Apply all parameter combinations in the hyperparameter space to the initial support vector machine, and use the accuracy, precision and other indicators to evaluate the initial support vector machine corresponding to each parameter combination. Select the parameter combination with the highest indicator as the optimal parameter combination, and use the corresponding initial support vector machine as the trained support vector machine, thus completing the construction of the support vector machine.

[0165] P4. Inputting the linear feature data into the logistic regression model, and outputting a first infection risk prediction value;

[0166] P5. Inputting the nonlinear feature data into the support vector machine, and outputting a second infection risk prediction value;

[0167] P6. According to the formula:

[0168] P final =η1P1+η2P2

[0169] Get the final infection risk prediction value P final . Among them, η1 and η2 represent weights, P1 represents the first infection risk prediction value, and P2 represents the second infection risk prediction value.

[0170] Since linear feature data is closely related to the infection risk of HIV or AIDS patients, it is easier to predict the infection risk. Therefore, when setting the weight value, the weight η1 needs to be greater than the weight η2.

[0171] In summary, the present invention can improve the robustness, stability and interpretability of the prediction model by introducing feature interaction and feature selection and reducing redundant features, more accurately predict the infection risk of HIV / AIDS patients, provide clinicians with more reliable prediction results, help formulate more effective treatment plans and early intervention measures, and reduce the infection risk of HIV / AIDS patients during treatment.

[0172] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An enhanced prediction method for HIV / AIDS patient infection risk, characterized in that: include: S1. Acquire clinical data of HIV patients or AIDS patients and perform preprocessing to obtain preprocessed clinical data; S2. Construct an enhanced infection risk prediction model; S3. Inputting the pre-processed clinical data into the enhanced infection risk prediction model, and outputting the infection risk prediction result of the HIV or AIDS patient.

2. An enhanced prediction method for HIV / AIDS patient infection risk based on claim 1, characterized in that: The clinical data of HIV patients or AIDS patients include demographic data, HIV-specific data, routine physiological data and influencing factors; the demographic data include the age, gender, race and geographical location of HIV patients or AIDS patients; the HIV-specific data include the CD4 cell count, disease load, medication record and efficacy of antiretroviral treatment, and previous history of opportunistic infections of HIV patients or AIDS patients; the disease load includes HIV level and RNA level; the routine physiological data include the body temperature, blood pressure, heart rate, routine blood data, biochemical indicators, and nutritional status of HIV patients or AIDS patients; the influencing factors include smoking, drinking, drug abuse history, socioeconomic status and mental and psychological status of HIV patients or AIDS patients; the routine blood data include white blood cell count, red blood cell count, and platelet count; The biochemical indices include liver function and kidney function; the nutritional status includes serum albumin level.

3. An enhanced prediction method for HIV / AIDS patient infection risk based on claim 2, characterized in that: The enhanced infection risk prediction model adopts a logistic regression model.

4. An enhanced prediction method for HIV / AIDS patient infection risk based on claim 3, characterized in that: The construction of the enhanced infection risk prediction model comprises the following steps: S2-1, based on the formula: Construct the first logistic regression model P0(Y=1|X); where Y represents the prediction result, X represents the feature data set, e represents the natural constant, β0 represents the bias term, β1 and βn represent the regression coefficients, and X1 and Xn represent the first feature data and the nth feature data respectively; S2-2, obtaining historical clinical data of HIV patients or AIDS patients and preprocessing them to obtain preprocessed historical clinical data; S2-3, inputting the pre-processed historical clinical data into the first logistic regression model, processing it through a recursive feature elimination algorithm, and obtaining corresponding selected feature data; S2-4, processing the acquired feature data by cross-validation technology and regularization method to obtain corresponding interactive feature data; S2-5, combining the interactive feature data and the preprocessed historical clinical data to obtain combined feature data; S2-6. Based on the merged feature data and the first logistic regression model, a second logistic regression model is constructed, and the corresponding historical infection risk prediction result is output; S2-7. Calculate the loss function corresponding to the second logistic regression model; S2-8. Based on the loss function, adjusting the regression coefficient of the second logistic regression model by a gradient descent method to obtain a third logistic regression model; S2-9. Repeat S2-6 to S2-8 until the preset number of iterations is reached to complete the construction of the enhanced infection risk prediction model.

5. An enhanced prediction method for HIV / AIDS patient infection risk based on claim 4, characterized in that: The pre-processed historical clinical data is input into the first logistic regression model, and processed by a recursive feature elimination algorithm to obtain corresponding selected feature data, comprising the following steps: S2-3-1, initializing the first logistic regression model; S2-3-2, inputting the pre-processed historical clinical data into the first logistic regression model to obtain corresponding regression coefficients; S2-3-3, selecting the pre-processed historical clinical data corresponding to the maximum regression coefficient, and adding it to the feature set as feature data; the initial number of features in the feature set is 0; S2-3-4, removing the characteristic data from the pre-processed historical clinical data to obtain updated historical clinical data; S2-3-5, calculating the p-value of the regression coefficient corresponding to the feature data added to the feature set; S2-3-6, determining whether the p value is less than a threshold value, if so, retaining the feature data to the feature set, and increasing the number of features in the feature set by 1; otherwise, removing the feature data from the feature set; S2-3-7. Input the updated historical clinical data into the first logistic regression model, repeat S2-3-2 to S2-3-6 until the number of feature data in the feature set reaches a preset feature number threshold, and obtain the selected feature data.

6. An enhanced prediction method for HIV / AIDS patient infection risk based on claim 5, characterized in that: The threshold in S2-3-6 is 0.

05.

7. An enhanced prediction method for HIV / AIDS patient infection risk based on claim 4, characterized in that: The formula of the second logistic regression model is: Wherein, σ(·) represents the sigmoid function, n represents the total number of feature data, βi represents the regression coefficient corresponding to the i-th merged feature data, βij represents the regression coefficient corresponding to the interactive feature corresponding to the i-th feature data and the j-th feature data, xi represents the i-th merged feature data, and xj represents the j-th merged feature data.

8. An enhanced prediction method for HIV / AIDS patient infection risk based on claim 4, characterized in that: The formula of the loss function Loss is: Among them, yi represents the label corresponding to the i-th historical infection risk prediction result, and P(yi|Xi) represents the i-th historical infection risk prediction result.

Images