A label differential privacy based average causal effect estimation method and system

Abstract

The application discloses a label differential privacy-based average causal effect estimation method and system. The application firstly trains a regression model to calculate the propensity score of each patient; then, the distance between different patients is calculated according to the propensity score of each patient, and the distance is sorted; the potential intervention result of each patient under another intervention condition is calculated based on the sorted distance matrix, noise is injected into the sum of the potential intervention results of all patients, and the estimation of the average causal effect is completed based on the noise-injected result; in addition, in the counterfactual estimation process, the maximum number of times that each patient is matched is limited, so that the influence of noise is reduced while the data accuracy is ensured.

Description

Technical Field

[0001] This invention relates to the field of data security technology, and in particular to a method and system for estimating average causal effects based on label differential privacy. Background Technology

[0002] Causal effect estimation aims to quantify the direct impact of an intervention on outcome variables, distinguishing intervention-induced changes from natural changes using scientific methods. This process is crucial for understanding causal relationships in complex systems. In the medical field, causal effect estimation based on patient data is often used to assist doctors' judgment. For example, analyzing whether a patient took medication and corresponding health indicators to assess the impact of medication use helps doctors make better decisions. However, using real intervention results for causal effect estimation may leak patients' privacy information. Therefore, perturbing intervention results before release during causal effect estimation to ensure the security of patients' privacy data is a problem worthy of in-depth research.

[0003] As a technology with rigorous theoretical safeguards, differential privacy technology introduces random noise into query results, making it impossible for attackers to infer information about a specific individual by comparing query results, even if they possess almost all other data. It is gaining increasing attention. However, directly applying differential privacy perturbations to the results can easily generate excessive noise, severely compromising data usability. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing average causal effect estimation techniques in the medical field, which either pose a threat to patient privacy or are prone to generating excessive noise, and to propose an average causal effect estimation method and system based on label differential privacy.

[0005] This invention overcomes the shortcomings of existing causal effect estimation techniques in the medical field mainly through the following two approaches: on the one hand, by perturbing the patient's intervention results to protect patient privacy; on the other hand, by limiting the maximum number of times each patient can be matched to reduce the sensitivity to noise injection.

[0006] The objective of this invention is achieved through the following technical solution: This invention discloses a method for estimating average causal effects based on label differential privacy, the method comprising: Step one: Train a regression model based on the patient's feature vector and intervention values, including the following sub-steps: 1.1 Normalize the patient's feature vector so that the value of the feature vector is between -1 and 1; 1.2 Using the normalized feature vectors as independent variables and the intervention values ​​as dependent variables, a logistic regression model is trained. 1.3 Based on the trained regression model, the output of the regression model is calculated using the patient's feature vector as input, and the output result is used as the propensity score for each patient; Step two involves grouping patients based on their intervention values, calculating the distance between each patient and all patients in the other group, and sorting them according to the distance. This includes the following sub-steps: 2.1 Patients with an intervention value of 1 were assigned to the treatment group, and patients with an intervention value of 0 were assigned to the control group; 2.2 For all patients in the treatment group, based on the patient's feature vector and propensity score, calculate the distance from each patient to all patients in the control group, and sort them in ascending order according to the distance, thus obtaining the distance sorting matrix from the treatment group patients to all patients in the control group; 2.3 For all patients in the control group, based on the patient’s feature vector and propensity score, calculate the distance from each patient to all patients in the treatment group, and sort them in ascending order according to the distance, thus obtaining the distance sorting matrix from the control group patients to all patients in the treatment group; Step 3: Based on the two distance ranking matrices and the actual intervention results for each patient, calculate the potential intervention results for each patient in the no-intervention scenario. Further calculate the sum of the potential intervention results for all patients when the intervention value is 1 and 0, including the following sub-steps: 3.1 Calculate the maximum matching limit for the processing group Maximum matching limit of control group ; 3.2 For all patients in the treatment group, based on their distance matrix to patients in the control group and combined with the maximum matching limit of the control group, calculate the potential intervention outcome for each patient when the intervention value is 0; 3.3 For all patients in the control group, the potential intervention outcome for each patient with an intervention value of 1 is calculated based on the distance matrix from the patients in the treatment group and the maximum matching limit of the treatment group. 3.4 Calculate the sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0; Step four involves perturbing the sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0 using the Laplace mechanism. The privacy budget allocated for this process is [missing information]. The average causal effect is estimated based on the perturbation value, including the following sub-steps: 4.1 The sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0 is perturbed by injecting Laplace noise; 4.2 The average causal effect is calculated based on two perturbation values ​​of the potential intervention outcome.

[0007] Furthermore, in step two, the distance metric function used is: In the formula Representing patients The tendency score, The regularization coefficient is . Representing patients eigenvectors, Representing patients The magnitude of the eigenvectors, This represents the dot product of two eigenvectors.

[0008] Furthermore, in step 3.1, the specific steps for calculating the maximum matching upper limit are as follows: a) Count the number of patients in the treatment group. Number of patients in the control group and calculate ; b) Set the number of matching neighbors And statistically analyze the distance of all samples in the ranking matrix. The maximum number of occurrences among neighboring patients ; c) If The maximum matching limit of the processing group is calculated according to the following formula. : In the formula Represents privacy budget, Represents the error coefficient. The number of patients in the control group The result is rounded to the nearest whole number. This represents taking the maximum value of the results within the parentheses, followed by calculating the maximum matching limit for the control group. : d) If The maximum matching limit of the control group is calculated according to the following formula. : In the formula Represents privacy budget, Represents the error coefficient. The number of patients in the treatment group The result is rounded to the nearest whole number. This represents taking the maximum value of the results within the parentheses, followed by calculating the maximum matching limit for the processing group. : Furthermore, in step 3.2, the specific steps for calculating the potential intervention outcomes for the treatment group patients are as follows: a) Initialize the number of matches for all patients in the control group to 0; b) Iterate through all patients in the treatment group sequentially, sorting the list according to each patient's distance from the control group. After excluding control group patients who have reached the maximum number of matches, select the patients with the smallest distance. Patients in the control group (using a set) (This indicates that the number of matches for these extracted control group patients is incremented by 1); c) Calculate the patient's condition according to the following formula. Potential intervention outcomes when the intervention value is 0 : In the formula Patients representing the control group Intervention results when the intervention value is 0.

[0009] Furthermore, in step 3.3, the specific steps for calculating the potential intervention outcomes for the control group patients are as follows: a) Initialize the number of matches for all patients in the initial treatment group to 0; b) Iterate through all patients in the control group sequentially, sorting the list according to each patient's distance from the treatment group. After excluding treatment group patients who have reached the maximum number of matches, select the patients with the smallest distance. Patients in each treatment group (using a set) (This indicates that the number of matches for these extracted treatment group patients is incremented by 1); c) Calculate the patient's condition according to the following formula. Potential intervention outcomes at an intervention value of 1 : In the formula Representative of the treatment group of patients Intervention results when the intervention value is 1.

[0010] Furthermore, in step 3.4, the specific steps for calculating the sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0 are as follows: a) Calculate the sum of potential intervention outcomes for all patients with an intervention value of 0, as follows: In the formula Represents the collection of all patients. Representing patients Potential intervention outcomes with an intervention value of 1; b) Calculate the sum of potential intervention outcomes for all patients with an intervention value of 0, as follows: In the formula Represents the collection of all patients. Representing patients Potential intervention outcomes with an intervention value of 0.

[0011] Furthermore, in step 4.1, the specific steps for injecting Laplace noise into the sum of potential intervention results are as follows: a) The perturbation values ​​for the sum of potential intervention outcomes when the intervention value is 1 are calculated as follows: In the formula Representing privacy budget Sensitivity is Laplace noise, where This represents global sensitivity, i.e., the maximum range of value variation (known value). b) The perturbation value for the sum of potential intervention outcomes when the intervention value is 0 is calculated as follows: In the formula Representing privacy budget Sensitivity is Laplace noise.

[0012] Furthermore, in step 4.2, the specific formula for calculating the average causal effect is as follows: Another aspect of the present invention discloses an average causal effect estimation system based on label differential privacy, the system comprising: Model training module: Trains a logistic regression model based on the patient's actual feature vectors and intervention values; Sample matching module: Based on the intervention value, patients are divided into treatment group and control group, the distance between each patient and all patients in the other group is calculated and sorted, thus obtaining two distance sorting matrices; The outcome estimation module calculates the potential intervention outcome for each patient in the uninterventional scenario based on two distance ranking matrices. It further calculates the sum of potential intervention outcomes for all patients when the intervention value is 1 and 0, and injects Laplace noise into the sum of potential intervention outcomes to perturb it. Finally, it achieves the average causal effect estimation, thereby reflecting the degree of influence of the intervention variable on the corresponding medical indicators of the intervention outcome.

[0013] The beneficial effects of this invention are as follows: 1. By perturbing the patient's intervention results while satisfying the privacy requirements of differential labeling, the patient's privacy information is effectively protected; 2. A limit was placed on the maximum number of matches for each patient to avoid introducing excessive noise; 3. To address the problem of estimating the average causal effect in the medical field, a label-based differential privacy estimation method is designed, which maximizes the usability of the average causal effect estimate while protecting patient privacy. Attached Figure Description

[0014] Figure 1 A flowchart of an average causal effect estimation method based on label differential privacy provided in an embodiment of the present invention; Figure 2 An example of normalized patient information provided in an embodiment of the present invention; Figure 3 A method based on the embodiments of the present invention is provided. Figure 2 Example of calculating potential intervention outcomes; Figure 4 This is a block diagram of an average causal effect estimation system based on label differential privacy, provided for an embodiment of the present invention. Detailed Implementation

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0017] like Figure 1 As shown in the figure, this embodiment provides an average causal effect estimation method based on label differential privacy. Based on the patient's feature vector, intervention value, and intervention result in the medical field, the average causal effect is calculated while satisfying the requirements of label differential privacy. This assesses the impact of specified intervention variables on the intervention result in the medical field, providing a reference for doctors' treatment. The specific steps are as follows: Step one: Train a regression model based on the patient's feature vector and intervention values, for example... Figure 2 As shown, the patient's feature vector includes age, height, gender, and heart rate. The intervention variable represents whether medication is being taken, with intervention values ​​including 1 (corresponding to medication) and 0 (corresponding to no medication). The intervention result corresponds to the patient's blood pressure (in millimeters of mercury), and includes the following sub-steps: 1.1 Normalize the patient's feature vector so that the value of the feature vector is between -1 and 1, for example... Figure 2This presents a summary of patient information containing 1000 patients and 4 feature vectors, with all feature vector values ​​normalized. 1.2 Using normalized feature vectors as independent variables and intervention values ​​as dependent variables, a logistic regression model is trained, for example... Figure 2 The four feature vectors are used as independent variables, and the intervention value consisting of 0 and 1 is used as the dependent variable. Based on this, a logistic regression model can be trained. 1.3 Based on the trained regression model, the output of the regression model is calculated using the patient's feature vector as input, and the output result is used as the propensity score for each patient regarding whether or not to take the medication. Step two involves grouping patients based on their intervention values, calculating the distance between each patient and all patients in the other group, and sorting them according to the distance. This includes the following sub-steps: 2.1 Patients with an intervention value of 1 were assigned to the treatment group, and patients with an intervention value of 0 were assigned to the control group. For example... Figure 2 Patients 0-399, the first 400 patients, were assigned to the control group, and patients 400-999, the last 600 patients, were assigned to the treatment group. 2.2 For all patients in the treatment group, based on the patient's feature vector and propensity score, calculate the distance from each patient to all patients in the control group, and sort them in ascending order according to the distance, thus obtaining the distance sorting matrix from the treatment group patients to all patients in the control group; 2.3 For all patients in the control group, based on the patient’s feature vector and propensity score, calculate the distance from each patient to all patients in the treatment group, and sort them in ascending order according to the distance, thus obtaining the distance sorting matrix from the control group patients to all patients in the treatment group; The specific details of the distance metric function are as follows: In the formula Representing patients The tendency score, This is the regularization coefficient (e.g., 0.1). Representing patients eigenvectors, Representing patients The magnitude of the eigenvectors, This represents the dot product of two eigenvectors.

[0018] Step 3: Based on the two distance ranking matrices and each patient's actual intervention outcome (i.e., whether or not medication was taken), calculate the potential intervention outcome (i.e., blood pressure value) for each patient in the no-intervention scenario. Further calculate the sum of the potential intervention outcomes for all patients when the intervention value is 1 and 0, including the following sub-steps: 3.1 Calculate the maximum matching limit for the processing group Maximum matching limit of control group ; The specific implementation details are as follows: a) Count the number of patients in the treatment group. Number of patients in the control group and calculate ,For example Figure 2 In the middle, it can be calculated that ; b) Set the number of matching neighbors (For example, set to 5), and count the number of patients in the top of the distance ranking matrix. The maximum number of occurrences among neighboring patients ; c) If The maximum matching limit of the processing group is calculated according to the following formula. : In the formula Represents privacy budget, This represents the error coefficient (set manually, for example, to 0.01). The number of patients in the control group The result is rounded to the nearest whole number. This represents taking the maximum value of the results within the parentheses, followed by calculating the maximum matching limit for the control group. : d) If The maximum matching limit of the control group is calculated according to the following formula. : In the formula Represents privacy budget, This represents the error coefficient (set manually, for example, to 0.01). The number of patients in the treatment group The result is rounded to the nearest whole number. This represents taking the maximum value of the results within the parentheses, followed by calculating the maximum matching limit for the processing group. : 3.2 For all patients in the treatment group, based on their distance matrix to patients in the control group and the maximum matching limit of the control group, calculate the potential intervention outcome for each patient when the intervention value is 0, for example... Figure 3 Example of the results in column 7 shown; The specific implementation details are as follows: a) Initialize the number of matches for all patients in the control group to 0; b) Iterate through all patients in the treatment group sequentially, sorting the list according to each patient's distance from the control group. After excluding control group patients who have reached the maximum number of matches, select the patients with the smallest distance. Patients in the control group (using a set) (This indicates that the number of matches for these extracted control group patients is incremented by 1); c) Calculate the patient's condition according to the following formula. Potential intervention outcomes when the intervention value is 0 : In the formula Patients representing the control group Intervention results when the intervention value is 0.

[0019] 3.3 For all patients in the control group, based on their distance matrix to patients in the treatment group and combined with the maximum matching upper limit of the treatment group, calculate the potential intervention outcome for each patient with an intervention value of 1, for example... Figure 3 Example of the results in column 8 shown; The specific implementation details are as follows: a) Initialize the number of matches for all patients in the initial treatment group to 0; b) Iterate through all patients in the control group sequentially, sorting the list according to each patient's distance from the treatment group. After excluding treatment group patients who have reached the maximum number of matches, select the patients with the smallest distance. Patients in each treatment group (using a set) (This indicates that the number of matches for these extracted treatment group patients is incremented by 1); c) Calculate the patient's condition according to the following formula. Potential intervention outcomes at an intervention value of 1 : In the formula Representative of the treatment group of patients Intervention results when the intervention value is 1.

[0020] 3.4 Calculate the sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0; The specific implementation details are as follows: a) Calculate the sum of potential intervention outcomes for all patients with an intervention value of 1 as follows: In the formula Represents the collection of all patients. Representing patients Among potential intervention outcomes with an intervention value of 1, such as for Figure 3Sum the values ​​in the column corresponding to the potential intervention outcome with an intervention value of 1 to obtain ; b) Calculate the sum of potential intervention outcomes for all patients with an intervention value of 0, as follows: In the formula Represents the collection of all patients. Representing patients In potential intervention outcomes with an intervention value of 0, such as for Figure 3 Sum the values ​​in the column corresponding to potential intervention outcomes with an intervention value of 0 to obtain... .

[0021] Step four involves perturbing the sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0 using the Laplace mechanism. The privacy budget allocated for this process is [missing information]. The average causal effect is estimated based on the perturbation value, including the following sub-steps: 4.1 The sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0 is perturbed by injecting Laplace noise; The specific implementation details are as follows: a) The perturbation values ​​for the sum of potential intervention outcomes when the intervention value is 1 are calculated as follows: In the formula Representing privacy budget (For example Sensitivity is Laplace noise, where Represents global sensitivity (e.g.) ), that is, the maximum range of variation of the value (known value), for example, for Figure 3 The sum of potential intervention outcomes with an intervention value of 1 is then noise-added to obtain... ; b) The perturbation value for the sum of potential intervention outcomes when the intervention value is 0 is calculated as follows: In the formula Representing privacy budget Sensitivity is Laplace noise, for example, for Figure 3 The sum of potential intervention outcomes when the intervention value is 0 is noise-added to obtain... .

[0022] 4.2 The average causal effect is calculated based on two perturbation values ​​of the potential intervention outcome; The specific implementation details are as follows: The specific formula for calculating the average causal effect is as follows: For example, for Figure 3 The final average causal effect was calculated as follows: The results indicate that patients taking medication experienced an average blood pressure reduction of 30 mmHg compared to those not taking medication, a finding that can provide important guidance for doctors in prescribing medication.

[0023] Another aspect of this invention discloses an average causal effect estimation system based on label differential privacy, such as... Figure 4 As shown, the system includes the following modules: Model training module: Trains a logistic regression model based on the patient's actual feature vectors and intervention values; Sample matching module: Based on the intervention value, patients are grouped (i.e., treatment group and control group), the distance between each patient and all patients in the other group is calculated, and the distances are sorted to obtain two distance sorting matrices; The outcome estimation module calculates the potential intervention outcome for each patient in the uninterventional scenario based on two distance ranking matrices. It further calculates the sum of potential intervention outcomes for all patients when the intervention value is 1 and 0, and injects Laplace noise into the sum of potential intervention outcomes to perturb it. Finally, it achieves the average causal effect estimation, thereby reflecting the degree of influence of the intervention variable on the corresponding medical indicators of the intervention outcome.

[0024] All relevant content of each step involved in the above method embodiments can be referenced from the functional description of the corresponding functional module, and will not be repeated here.

[0025] This invention designs an average causal effect estimation method and system based on label differential privacy. By limiting the maximum number of times each patient can be matched during the calculation process, the availability of average causal effect estimation is improved as much as possible while protecting the privacy of patient intervention results.

[0026] The above description is merely a preferred embodiment of the present invention. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the technical solutions of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall still fall within the protection scope of the technical solutions of the present invention.

Claims

1. A method for estimating average causal effects based on label-based differential privacy, characterized in that the method includes: Step one: Train a regression model based on the patient's feature vector and intervention values, including the following sub-steps: 1.1 Normalize the patient's feature vector so that the value of the feature vector is between -1 and 1; 1.2 A logistic regression model is trained by using the normalized feature vector as the independent variable and the intervention value as the dependent variable; 1.3 Based on the trained regression model, the output of the regression model is calculated using the patient's feature vector as input, and the output result is used as the propensity score for each patient; Step two involves grouping patients based on their intervention values, calculating the distance between each patient and all patients in the other group, and sorting them according to the distance. This includes the following sub-steps: 2.1 Patients with an intervention value of 1 were assigned to the treatment group, and patients with an intervention value of 0 were assigned to the control group; 2.2 For all patients in the treatment group, based on the patient's feature vector, calculate the distance from each patient to all patients in the control group, and sort them in ascending order according to the distance, thus obtaining the distance sorting matrix from the treatment group patients to all patients in the control group; 2.3 For all patients in the control group, based on the patient's feature vector, calculate the distance from each patient to all patients in the treatment group, and sort them in ascending order according to the distance, thus obtaining the distance sorting matrix from the control group patients to all patients in the treatment group; Step 3: Based on the two distance ranking matrices and the actual intervention results for each patient, calculate the potential intervention results for each patient in the no-intervention scenario. Further calculate the sum of the potential intervention results for all patients when the intervention value is 1 and 0, including the following sub-steps: 3.1 Calculate the maximum matching limit for the processing group Maximum matching limit with control group Specifically: a) Count the number of patients in the treatment group. Number of patients in the control group and calculate ; b) Set the number of matching neighbors And statistically analyze the distance ranking matrix of all patients. The maximum number of occurrences among neighboring patients ; c) If The maximum matching limit of the processing group is calculated according to the following formula. : In the formula Represents privacy budget, Represents the error coefficient. The number of patients in the control group The result is rounded to the nearest whole number. This represents taking the maximum value of the results within the parentheses, followed by calculating the maximum matching limit for the control group. : d) If The maximum matching limit of the control group is calculated according to the following formula. : In the formula The number of patients in the treatment group is then used to calculate the maximum matching limit for the treatment group. : 3.2 For all patients in the treatment group, based on their distance matrix to patients in the control group and combined with the maximum matching limit of the control group, calculate the potential intervention outcome for each patient when the intervention value is 0; 3.3 For all patients in the control group, the potential intervention outcome for each patient with an intervention value of 1 is calculated based on the distance matrix from the patients in the treatment group and the maximum matching limit of the treatment group. 3.4 Calculate the sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0; Step four: The sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0 is perturbed using the Laplace mechanism, and the average causal effect is estimated based on the perturbation values.

2. The method for estimating the average causal effect based on label-based differential privacy according to claim 1, characterized in that, in step two, the distance metric function used is: In the formula Representing patients respectively The tendency score, The regularization coefficient is . Representing patients respectively eigenvectors, Representing patients respectively The magnitude of the eigenvectors, This represents the dot product of two eigenvectors.

3. The method for estimating the average causal effect based on label differential privacy according to claim 1, characterized in that, In step 3.2, the specific steps for calculating the potential intervention outcomes for the treatment group patients are as follows: a) Initialize the number of matches for all patients in the control group to 0; b) Iterate through all patients in the treatment group sequentially, sorting the list according to each patient's distance from the control group. After excluding control group patients who have reached the maximum number of matches, select the patients with the smallest distance. Patients in the control group, using a set This indicates that the number of matches for these extracted control group patients will be incremented by 1; c) Calculate the patient's condition according to the following formula. Potential intervention outcomes when the intervention value is 0 : In the formula Patients representing the control group Intervention results when the intervention value is 0.

4. The method for estimating the average causal effect based on label differential privacy according to claim 1, characterized in that, In step 3.3, the specific steps for calculating the potential intervention outcomes for patients in the control group are as follows: a) Initialize the number of matches for all patients in the initial treatment group to 0; b) Iterate through all patients in the control group sequentially, sorting the list according to each patient's distance from the treatment group. After excluding treatment group patients who have reached the maximum number of matches, select the patients with the smallest distance. Patients in a treatment group, using a set This indicates that the number of matches for these extracted treatment group patients will be incremented by 1; c) Calculate the patient's condition according to the following formula. Potential intervention outcomes at an intervention value of 1 : In the formula Representative of the treatment group of patients Intervention results when the intervention value is 1.

5. The method for estimating the average causal effect based on label-based differential privacy according to claim 1, characterized in that, In step 3.4, the specific steps for calculating the sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0 are as follows: a) Calculate the sum of potential intervention outcomes for all patients with an intervention value of 1, as follows: In the formula Represents the collection of all patients. Representing patients Potential intervention outcomes with an intervention value of 1; b) Calculate the sum of potential intervention outcomes for all patients with an intervention value of 0, as follows: In the formula Representing patients Potential intervention outcomes with an intervention value of 0.

6. The method for estimating the average causal effect based on label-based differential privacy according to claim 1, characterized in that, In step four, the sum of potential intervention outcomes for all patients at intervention values ​​of 1 and 0 is perturbed by injecting Laplace noise. The specific steps are as follows: a) Calculate the sum of potential intervention outcomes when the intervention value is 1. The disturbance values ​​are as follows: In the formula Representing privacy budget Sensitivity is Laplace noise, where Represents global sensitivity. This represents the maximum matching limit for the processing group; b) Calculate the sum of potential intervention outcomes when the intervention value is 0. The disturbance values ​​are as follows: In the formula Representing privacy budget Sensitivity is Laplace noise, This represents the maximum matching limit for the control group.

7. The method for estimating the average causal effect based on label-based differential privacy according to claim 1, characterized in that, In step four, the average causal effect The calculation formula is: In the formula These are the perturbation values ​​of the sum of potential intervention outcomes when the intervention value is 1 and 0, respectively. The number of patients in the treatment group. The number of patients in the control group.

8. A label-based differential privacy-based average causal effect estimation system, characterized in that, The system, employing the average causal effect estimation method based on label differential privacy as described in any one of claims 1-7, comprises: Model training module: Trains a logistic regression model based on the patient's actual feature vectors and intervention values; Sample matching module: Based on the intervention value, patients are divided into treatment group and control group, the distance between each patient and all patients in the other group is calculated and sorted, thus obtaining two distance sorting matrices; The outcome estimation module calculates the potential intervention outcome for each patient in the uninterventional scenario based on two distance ranking matrices. It further calculates the sum of potential intervention outcomes for all patients when the intervention value is 1 and 0, and injects Laplace noise into the sum of potential intervention outcomes to perturb it. Finally, it achieves the average causal effect estimation, thereby reflecting the degree of influence of the intervention variable on the corresponding medical indicators of the intervention outcome.

Images