A method for predicting disease decline rate using mixed effects models

Abstract

The present invention relates to a method for predicting a disease decline rate through a mixed effect model, and belongs to the field of medical data analysis. A method for predicting a disease decline rate through a mixed effect model comprises the following steps: S1, inputting basic information and time series detection data of a patient; S2, calculating the decline rate of the HBsAg level of the patient at each detection moment relative to the baseline; S3, establishing a group trajectory model, using a conditional probability formula to calculate the probability that the patient belongs to different heterogeneous latent category groups, and judging the heterogeneous latent category group to which the patient is most likely to belong; S4, establishing a time effect prediction model, using the patient's HBsAg level decline rate data to predict the patient's future HBsAg level decline trend; S5, coupling the time effect prediction model with the group trajectory model to improve the prediction accuracy of the patient's future HBsAg level decline rate. By combining the time effect prediction model with the group trajectory model to predict the decline rate, the individual differences of the patient and the dynamic changes of time are considered, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical data analysis, and in particular to a method for predicting disease decline rate through a mixed effect model. Background Art

[0002] Hepatitis B, as a global health problem, affects the quality of life of hundreds of millions of people. HBsAg (hepatitis B surface antigen) is a key biomarker for hepatitis B virus infection. The rate of decline of HBsAg levels relative to baseline at different time points during treatment not only reflects the patient's clinical course, but is also closely related to the timely adjustment of treatment strategies and the ideal clinical response outcome. Most of the existing studies on HBsAg levels focus on dynamic evaluation, and reliable predictors of the disappearance or decrease of HBsAg levels have not yet been established. There are only a few studies on the prediction of HBsAg dynamic levels in patients with inactive CHB. Therefore, clinical decision support urgently needs to develop a real-time prediction model for the rate of decline of HBsAg levels relative to baseline at each treatment moment to help patients achieve better functional cure clinical outcomes.

[0003] The purpose of the present invention is to develop a statistical prediction model that can better predict the decline rate of HBsAg level relative to the baseline of patients during treatment in real time. Specifically, the group trajectory model is first used to identify different potential categories in the heterogeneous patient population, and then the mixed effect regression technology is used to obtain the time effect prediction model of the decline rate of HBsAg level at each treatment moment. Finally, the group trajectory model is coupled in the mixed effect regression to obtain a new time plus group dual effect prediction model to predict the decline rate of HBsAg level relative to the baseline of the patient in the future, thereby improving the prediction accuracy. Summary of the invention

[0004] Based on this, it is necessary to provide a method for predicting the disease decline rate through a mixed effects model to address the limitations of existing methods for predicting the decline rate of HBsAg levels.

[0005] A method for predicting disease decline rate by mixed effect model comprises the following steps:

[0006] S1. Input the patient's basic information and time series test data;

[0007] S2. Calculate the decline rate of the patient's HBsAg level relative to the baseline at each testing moment based on the patient's time series test data;

[0008] S3. Establish a group trajectory model, use the conditional probability formula to calculate the probability that the patient belongs to different heterogeneous latent class groups, and determine the heterogeneous latent class group to which the patient is most likely to belong;

[0009] S4. Establish a time effect prediction model and use the patient's HBsAg level decline rate data to predict the patient's future HBsAg level decline trend;

[0010] S5. Couple the time effect prediction model and the group trajectory model to improve the prediction accuracy of the patient's future HBsAg level decline rate.

[0011] The present application discloses a method for predicting the disease decline rate through a mixed effect model, which provides a personalized HBsAg level decline rate prediction for each patient based on the patient's basic information and time series detection data, thereby more accurately reflecting individual differences. A group trajectory model is established, and the conditional probability formula is used to calculate the probability that the patient belongs to different heterogeneous latent category groups, and the heterogeneous latent category group to which the patient is most likely to belong is determined, which helps doctors better understand the disease progression patterns of different patients and provide more targeted treatment suggestions for patients in different groups. Through the time effect prediction model, the patient's future HBsAg level change trend can be predicted, which helps doctors understand the patient's condition changes in advance and provide a basis for treatment adjustments. Combining the time effect prediction model with the group trajectory model can more accurately predict the patient's future HBsAg level decline rate. This method of predicting the future HBsAg level decline rate using a mixed effect model takes into account the individual differences of patients and the dynamic changes of time, and improves the accuracy and reliability of the prediction.

[0012] In any of the above technical solutions, in step S2, the decline rate is calculated by the following formula:

[0013]

[0014] in, represents the baseline HBsAg level, represents the HBsAg level at the tth detection time week, It represents the decrease rate of HBsAg level of patient i at the tth detection time week relative to the baseline.

[0015] By introducing the baseline HBsAg level and the HBsAg level at the tth detection moment, the decline rate of the disease is calculated, making the prediction results closer to the actual situation. Predicting the HBsAg level of patients at specific time points can formulate individualized treatment plans for patients, optimize treatment effects, reduce unnecessary drug use and side effects, and timely predict the decline rate of the disease, helping doctors to identify changes in the condition early and take appropriate intervention measures. Due to the use of a linear mixed effects model, a reasonable adjustment mechanism can be set after the prediction to avoid the prediction results not being in line with the actual situation.

[0016] In any of the above technical solutions, in step S3, the heterogeneous latent category groups include a fast high clearance group, a delayed high clearance group and a slow low clearance group.

[0017] By dividing patients into different heterogeneous latent class groups, the characteristics of disease decline rates in different patient groups can be captured more accurately. This grouping method helps to identify the specific needs and responses of different patient groups, thereby improving the accuracy of the prediction model. At the same time, by dividing different latent class groups, more information about the disease process can be obtained. For example, the fast high clearance group may represent individuals with rapid disease recovery, while the slow low clearance group may represent individuals with slower disease recovery. This information is important for developing targeted treatment strategies and evaluating treatment effects.

[0018] In any of the above technical solutions, in step S3,

[0019] For patient i, the detection time point is w i,0 ,w i,1 ,w i,2 ,…,w i,t , where w i,0 represents the time of hospitalization and treatment, recorded as 0 week, w i,1 Indicates the number of weeks after the first test after hospitalization, w i,2 Indicates the number of weeks after hospitalization for the second test, w i,t represents the number of weeks after patient i was admitted to the hospital at the time of the tth test, w i,0 ,w i,1 ,w i,2 ,…,w i,t It is not necessarily a continuous positive integer; the actual decrease rate of HBsAg level relative to the baseline at each detection time week is Calculated by the following formula:

[0020]

[0021] in, represents the baseline HBsAg level, represents the HBsAg level at the tth detection time week, It represents the decline rate of HBsAg level of patient i at the tth detection time week relative to the baseline;

[0022] according to

[0023]

[0024] The predicted decrease rate of HBsAg level relative to the baseline at each detection time week of the patient i when he belongs to different heterogeneous latent class groups is calculated and recorded as and

[0025] Let T = count{w i,1 ,w i,2 ,…,w i,t}, represents the number of detection time points other than the baseline detection time, according to the conditional probability formula:

[0026]

[0027] The probability value of patient i belonging to different heterogeneous latent class groups is calculated, where σ1=0.17052475, σ2=0.28058448, σ3=0.30667541, y i,t represents the decline rate of HBsAg level relative to the baseline of patient i at the tth detection time, is the predicted decline rate of HBsAg level relative to the baseline of patient i belonging to trajectory group j at the tth detection week, Φ(x) is the standard normal cumulative distribution function, and its calculation formula is

[0028] Compare the three conditional probability values ​​P(j|Y i ),j=1,2,3, where patient i belongs to the heterogeneous latent class group corresponding to the maximum conditional probability value.

[0029] By considering the actual decline rate of HBsAg levels relative to the baseline at different testing time weeks, it is possible to provide each patient with a personalized disease decline rate prediction, thereby more accurately evaluating the treatment effect and disease progression. The conditional probability formula is used to calculate the probability value of patients belonging to different heterogeneous latent class groups to determine the heterogeneous latent class group to which the patient belongs, which is helpful for further studying the disease characteristics and treatment responses of different groups and providing support for basic research and clinical application of the disease. This method allows patients to have discontinuous testing time points, making full use of the patient's test data and improving the efficiency of data utilization.

[0030] In any of the above technical solutions, in step S4,

[0031] Using the data of the HBsAg level decline rate of patient i from week 0 to week 24, with week 0 at the time of initial diagnosis as the baseline week, the time effect prediction model of the relative baseline decline rate of HBsAg level in different detection weeks of patients is:

[0032]

[0033] in, It represents the relative decrease rate of HBsAg level of patient i at the tth detection time week after admission to hospital under the time effect prediction model; is the patient's baseline week; is the number of weeks that patient i participates in the prediction, that is, w i,t The most recent detection week of patient i before week; For patient i in The actual decline rate of HBsAg level relative to the baseline in the week; Indicates the forecast time interval in weeks; represents the time effect intercept at different prediction time intervals; x 1,i Indicates gender, male is 1, female is 0; x 2,i represents the patient's age, x 3,i represents the patient's BMI value, represents the baseline HBsAg level of patient i, represents the baseline ALT level of patient i; x 6,i represents the baseline HBeAg level of patient i. If the baseline HBeAg test value is greater than 1, then x 6,i =1, indicating HBeAg positive, otherwise x 6,i =0, indicating HBeAg negative; x 7,i represents the baseline HBVDNA level of patient i. If the baseline HBVDNA detection value is greater than the detection limit, then x 7,i =1, indicating viremia, otherwise x 7,i =0, indicating non-viremia.

[0034] By introducing multiple influencing factors, such as gender, age, BMI, baseline HBsAg level, baseline ALT level, baseline HBeAg level, and baseline HBVDNA level, the time effect prediction model can provide personalized disease decline rate prediction for each patient. The time effect prediction model introduces the time effect, that is, the time effect intercept of different prediction time intervals, so that the model can capture the trend of disease decline rate over time and improve the accuracy of prediction.

[0035] In any of the above technical solutions, when the number of weeks to be predicted does not exceed 24 weeks, that is, w i,t ≤24, baseline weeks The time-effect prediction model was directly used to predict the decline rate of HBsAg level relative to the baseline;

[0036] Time effect intercept The maximum number of weeks is only 24 weeks. When the number of weeks to be predicted exceeds 24 weeks, that is, w i,t >24, the time effect prediction model was corrected by sliding the time window; except for the baseline week, the time window length of the data used in the construction of the time effect prediction model was 24 weeks, and the time window length was set to 24 weeks;

[0037] When the number of weeks to be predicted is w i,t >24, keep the time window length as 24 weeks and slide the time window from the time of the first diagnosis in steps of 1 week until the sliding time window just contains the week number w to be predicted i,t ; At this time, the baseline week is re-established for the sliding time window, that is, Week is the baseline week of the sliding time window; thus, the time effect prediction model is continued to be used to predict the wth week. i,t The rate of decrease of HBsAg level relative to baseline in 1 week.

[0038] When the number of weeks to be predicted does not exceed 24 weeks, the time effect prediction model can be directly used to predict the decline rate of HBsAg level relative to the baseline. Since the maximum number of weeks of the time effect intercept is only 24 weeks, this method has a high accuracy in predicting the disease decline rate in the short term, and is also very helpful for quickly evaluating the treatment effect and adjusting the treatment plan in time. Setting the time window length to 24 weeks can make full use of existing data, while avoiding the data sparsity problem caused by a long time span, and improving the prediction accuracy of the time effect prediction model. When the number of weeks to be predicted exceeds 24 weeks, the time effect prediction model is corrected using the time window sliding idea, keeping the window length at 24 weeks to ensure the continued effectiveness of the prediction model, and re-establishing the baseline week during the sliding time window, that is, taking the baseline week of the sliding time window as the new prediction starting point. This approach of dynamically adjusting the prediction baseline can better reflect the real-time changes in the disease state, thereby improving the accuracy of the prediction.

[0039] In any of the above technical solutions, the HBsAg level decline rate data of the patient from week 0 to week 24 is used, and the 0 week at the time of initial diagnosis is taken as the baseline week. The time effect prediction model and the group trajectory model are coupled to construct the following time-plus-group dual-effect prediction model of the HBsAg level decline rate relative to the baseline based on the group trajectory model:

[0040]

[0041] in, It represents the relative decrease rate of HBsAg level of patient i at the tth detection time week after admission under the time plus group double effect prediction model; is the patient's baseline week; The number of weeks that the patient will participate in the forecast, that is, w i,t The most recent test week for the patient before the week; For this patient The actual decline rate of HBsAg relative to the baseline at each time week; is the number of weeks for the forecast interval; Indicates that the patient is in wi,t The time effect intercept when the weekly forecast is the jth latent class group, j = 1, 2, 3; x 1,i Indicates gender, male is 1, female is 0; x 2,i represents the patient's age, x 3,i represents the patient's BMI value, represents the baseline HBsAg level of patient i, represents the baseline ALT level of patient i; x 6,i represents the baseline HBeAg level of patient i. If the baseline HBeAg test value is greater than 1, then x 6,i =1, indicating HBeAg positive, otherwise x 6,i =0, indicating HBeAg negative; x 7,i represents the baseline HBVDNA level of patient i. If the baseline HBVDNA detection value is greater than the detection limit, then x 7,i =1, indicating viremia, otherwise x 7,i =0, indicating non-viremia.

[0042] By coupling the time effect prediction model with the group trajectory model, the time factor and the heterogeneity of the patient population are considered at the same time, so as to improve the accuracy of the prediction of the future HBsAg level decline rate. Using the HBsAg level decline rate data of patients from 0 to 24 weeks, the prediction can be dynamically updated according to the patient's latest test results, making the prediction results closer to the patient's actual condition. By introducing the concept of latent category groups, patient groups with similar disease decline trends can be identified, which helps doctors adopt more targeted treatment strategies for different patient groups.

[0043] In any of the above technical solutions, in step S5,

[0044] When the number of weeks to be forecasted does not exceed 24, the following steps are taken:

[0045] S51. Determine the baseline week and select the starting week of the initial diagnosis and treatment or study as the baseline week;

[0046] S52. Based on the baseline week data and the time effect prediction model, directly predict the decline rate of HBsAg level relative to the baseline in the next no more than 24 weeks.

[0047] By determining the baseline week and selecting the starting week of the initial diagnosis or study as the baseline, the initial state of disease development can be captured more accurately, providing a more reliable starting point for subsequent predictions. When the number of weeks to be predicted does not exceed 24 weeks, predictions are made directly based on the baseline week data and the time effect prediction model, avoiding complex data processing and model adjustments and improving the efficiency of predictions. This makes this method suitable for predicting the disease decline rate in the short term, which is of great significance for clinical decision-making or research that requires a quick understanding of the disease development trend.

[0048] In any of the above technical solutions, in step S5,

[0049] When the number of weeks to be forecasted exceeds 24, the following steps are taken:

[0050] S53, using a time window sliding method to calibrate the time effect prediction model;

[0051] S54, setting the time window length. Except for the baseline week, the data time window length used in constructing the time effect prediction model is 24;

[0052] S55, when making a prediction, keep the time window length at 24 weeks, and slide the time window from the time of the first diagnosis in steps of 1 week. As time goes by, the time window will include the number of weeks to be predicted;

[0053] S56, re-establishing the baseline week, and taking the starting week of the sliding time window as the new baseline week after each sliding time window;

[0054] S57. Based on the data of the sliding time window and the established time effect prediction model, predict the future decline rate of HBsAg level relative to the baseline.

[0055] When the prediction period exceeds 24 weeks, the time effect prediction model is corrected by sliding the time window. This method allows the time effect prediction model to dynamically adjust according to the passage of time to better adapt to the needs of long-term prediction, thereby improving the accuracy and practicality of the prediction model. By sliding the time window, this method can include the number of weeks to be predicted over time, thereby realizing the prediction of long time periods, so that the method can adapt to the prediction needs of different lengths and has a wider range of applications. After each sliding time window, the baseline week is re-established, and the starting week of the sliding time window is used as the baseline week. The flexible baseline adjustment method enables the time effect prediction model to adapt to data changes in different time periods, further improving the accuracy of the prediction. This method is based on the data of the sliding time window and the established time effect prediction model to predict the future decline rate of HBsAg levels relative to the baseline. This prediction method based on real-time data can obtain prediction results more quickly and improve work efficiency.

[0056] In any of the above technical solutions, in step S1, the basic information includes gender, age and BMI, and the time series detection data includes detection date, HBsAg level, HBeAg level, ALT level and HBV DNA level.

[0057] By considering the patient's gender, age, BMI and other factors, it can provide each patient with a personalized disease decline rate prediction, which helps doctors develop more personalized treatment plans. By collecting time series detection data, it can monitor the patient's condition in real time, promptly detect signs of worsening or improvement, and improve prediction accuracy, providing doctors with timely feedback information to adjust treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A flowchart of the steps in the method for predicting disease decline rate using mixed effects models;

[0059] Figure 2 A flowchart of the steps for predicting the decline rate when the number of weeks to be predicted does not exceed 24 weeks;

[0060] Figure 3 Flow chart of the steps for predicting the decline rate when the number of weeks to be predicted exceeds 24 weeks. DETAILED DESCRIPTION

[0061] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0062] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0063] A method for predicting disease decline rate by using a mixed effects model according to some embodiments of the present invention will be described below with reference to the accompanying drawings.

[0064] Example 1

[0065] like Figure 1 As shown, this embodiment discloses a method for predicting disease decline rate by using a mixed effect model, characterized in that it includes the following steps:

[0066] S1. Input the patient's basic information and time series test data;

[0067] S2. Calculate the decline rate of the patient's HBsAg level relative to the baseline at each testing moment based on the patient's time series test data;

[0068] S3. Establish a group trajectory model, use the conditional probability formula to calculate the probability that the patient belongs to different heterogeneous latent class groups, and determine the heterogeneous latent class group to which the patient is most likely to belong;

[0069] S4. Establish a time effect prediction model and use the patient's HBsAg level decline rate data to predict the patient's future HBsAg level decline trend;

[0070] S5. Couple the time effect prediction model and the group trajectory model to improve the prediction accuracy of the patient's future HBsAg level decline rate.

[0071] The present application discloses a method for predicting the disease decline rate through a mixed effect model, which provides a personalized HBsAg level decline rate prediction for each patient based on the patient's basic information and time series detection data, thereby more accurately reflecting individual differences. A group trajectory model is established, and the conditional probability formula is used to calculate the probability that the patient belongs to different heterogeneous latent category groups, and the heterogeneous latent category group to which the patient is most likely to belong is determined, which helps doctors better understand the disease progression patterns of different patients and provide more targeted treatment suggestions for patients in different groups. Through the time effect prediction model, the patient's future HBsAg level change trend can be predicted, which helps doctors understand the patient's condition changes in advance and provide a basis for treatment adjustments. Combining the time effect prediction model with the group trajectory model can more accurately predict the patient's future HBsAg level decline rate. This method of predicting the future HBsAg level decline rate using a mixed effect model takes into account the individual differences of patients and the dynamic changes of time, and improves the accuracy and reliability of the prediction.

[0072] In any of the above technical solutions, in step S2, the descent rate is calculated by the following formula:

[0073]

[0074] in, represents the baseline HBsAg level, represents the HBsAg level at the tth detection time week, It represents the decrease rate of HBsAg level of patient i at the tth detection time week relative to the baseline.

[0075] By introducing the baseline HBsAg level and the HBsAg level at the tth detection moment, the decline rate of the disease is calculated, making the prediction results closer to the actual situation. Predicting the HBsAg level of patients at specific time points can formulate individualized treatment plans for patients, optimize treatment effects, reduce unnecessary drug use and side effects, and timely predict the decline rate of the disease, helping doctors to identify changes in the condition early and take appropriate intervention measures. Due to the use of a linear mixed effects model, a reasonable adjustment mechanism can be set after the prediction to avoid the prediction results not being in line with the actual situation.

[0076] In any of the above technical solutions, in step S3, the heterogeneous latent category groups include a fast high clearance group, a delayed high clearance group and a slow low clearance group.

[0077] By dividing patients into different heterogeneous latent class groups, the characteristics of disease decline rates in different patient groups can be captured more accurately. This grouping method helps to identify the specific needs and responses of different patient groups, thereby improving the accuracy of the prediction model. At the same time, by dividing different latent class groups, more information about the disease process can be obtained. For example, the fast high clearance group may represent individuals with rapid disease recovery, while the slow low clearance group may represent individuals with slower disease recovery. This information is important for developing targeted treatment strategies and evaluating treatment effects.

[0078] In any of the above technical solutions, in step S3,

[0079] For patient i, the detection time point is w i,0 ,w i,1 ,w i,2 ,…,w i,t , where w i,0 represents the time of hospitalization and treatment, recorded as 0 week, w i,1 Indicates the number of weeks after the first test after hospitalization, w i,2 Indicates the number of weeks after hospitalization for the second test, w i,t represents the number of weeks after patient i was admitted to the hospital at the time of the tth test, w i,0 ,w i,1 ,w i,2 ,…,w i,t It is not necessarily a continuous positive integer; the actual decrease rate of HBsAg level relative to the baseline at each detection time week is Calculated by the following formula:

[0080]

[0081] in, represents the baseline HBsAg level, represents the HBsAg level at the tth detection time week, It represents the decline rate of HBsAg level of patient i at the tth detection time week relative to the baseline;

[0082] according to

[0083]

[0084] The predicted decrease rate of HBsAg level relative to the baseline at each detection time week of the patient i when he belongs to different heterogeneous latent class groups is calculated and recorded as and

[0085] Let T = count{w i,1 ,w i,2 ,…,w i,t}, represents the number of detection time points other than the baseline detection time, according to the conditional probability formula:

[0086]

[0087] The probability value of patient i belonging to different heterogeneous latent class groups is calculated, where σ1=0.17052475, σ2=0.28058448, σ3=0.30667541, y i,t represents the decline rate of HBsAg level relative to the baseline of patient i at the tth detection time, is the predicted decline rate of HBsAg level relative to the baseline of patient i belonging to trajectory group j at the tth detection week, Φ(x) is the standard normal cumulative distribution function, and its calculation formula is

[0088] Compare the three conditional probability values ​​P(j|Y i ),j=1,2,3, where patient i belongs to the heterogeneous latent class group corresponding to the maximum conditional probability value.

[0089] By considering the actual decline rate of HBsAg levels relative to the baseline at different testing time weeks, it is possible to provide each patient with a personalized disease decline rate prediction, thereby more accurately evaluating the treatment effect and disease progression. The conditional probability formula is used to calculate the probability value of patients belonging to different heterogeneous latent class groups to determine the heterogeneous latent class group to which the patient belongs, which is helpful for further studying the disease characteristics and treatment responses of different groups and providing support for basic research and clinical application of the disease. This method allows patients to have discontinuous testing time points, making full use of the patient's test data and improving the efficiency of data utilization.

[0090] In any of the above technical solutions, in step S4,

[0091] Using the data of the HBsAg level decline rate of patient i from week 0 to week 24, with week 0 at the time of initial diagnosis as the baseline week, the time effect prediction model of the relative baseline decline rate of HBsAg level in different detection weeks of patients is:

[0092]

[0093] in, It represents the relative decrease rate of HBsAg level of patient i at the tth detection time week after admission to hospital under the time effect prediction model; is the patient's baseline week; is the number of weeks that patient i participates in the prediction, that is, w i,t The most recent detection week of patient i before week; For patient i in The actual decline rate of HBsAg level relative to the baseline in the week; Indicates the forecast time interval in weeks; represents the time effect intercept of different prediction time intervals, and its values ​​are shown in Table 1; x 1,i Indicates gender, male is 1, female is 0; x 2,i represents the patient's age, x 3,i represents the patient's BMI value, represents the baseline HBsAg level of patient i, represents the baseline ALT level of patient i; x 6,i represents the baseline HBeAg level of patient i. If the baseline HBeAg test value is greater than 1, then x 6,i =1, indicating HBeAg positive, otherwise x 6,i =0, indicating HBeAg negative; x 7,i represents the baseline HBVDNA level of patient i. If the baseline HBVDNA detection value is greater than the detection limit, then x 7,i =1, indicating viremia, otherwise x 7,i =0, indicating non-viremia.

[0094] Table 1. Time effect intercepts at different prediction time intervals in the time effect prediction model

[0095] Week intercept Week intercept Week intercept 1 0.329907 9 0.699365 17 0.915777 2 0.384212 10 0.737616 18 0.925492 3 0.432815 11 0.772243 19 0.940163 4 0.482821 12 0.799119 20 0.943000 5 0.532025 13 0.830567 21 0.957224 6 0.579139 14 0.853614 22 0.968566 7 0.622765 15 0.881314 23 0.964721 8 0.664618 16 0.891146 24 0.937522

[0096] By introducing multiple influencing factors, such as gender, age, BMI, baseline HBsAg level, baseline ALT level, baseline HBeAg level and baseline HBV DNA level, the time effect prediction model can provide personalized disease decline rate prediction for each patient. The time effect prediction model introduces the time effect, that is, the time effect intercept of different prediction time intervals, so that the model can capture the trend of disease decline rate over time and improve the accuracy of prediction.

[0097] In any of the above technical solutions, when the number of weeks to be predicted does not exceed 24 weeks, that is, w i，t ≤24, baseline weeks The time-effect prediction model was directly used to predict the decline rate of HBsAg level relative to the baseline;

[0098] Time effect intercept The maximum number of weeks is only 24 weeks. When the number of weeks to be predicted exceeds 24 weeks, that is, w i，t >24, the time effect prediction model was corrected by sliding the time window; except for the baseline week, the time window length of the data used in the construction of the time effect prediction model was 24 weeks, and the time window length was set to 24 weeks;

[0099] When the number of weeks to be predicted is w i，t >24, keep the time window length as 24 weeks and slide the time window from the time of the first diagnosis in steps of 1 week until the sliding time window just contains the week number w to be predicted i，t ; At this time, the baseline week is re-established for the sliding time window, that is, Week is the baseline week of the sliding time window; thus, the time effect prediction model is continued to be used to predict the wth week. i，t The rate of decrease of HBsAg level relative to baseline in 1 week.

[0100] When the number of weeks to be predicted does not exceed 24 weeks, the time effect prediction model can be directly used to predict the decline rate of HBsAg level relative to the baseline. Since the maximum number of weeks of the time effect intercept is only 24 weeks, this method has a high accuracy in predicting the disease decline rate in the short term, and is also very helpful for quickly evaluating the treatment effect and adjusting the treatment plan in time. Setting the time window length to 24 weeks can make full use of existing data, while avoiding the data sparsity problem caused by a long time span, and improving the prediction accuracy of the time effect prediction model. When the number of weeks to be predicted exceeds 24 weeks, the time effect prediction model is corrected using the time window sliding idea, keeping the window length at 24 weeks to ensure the continued effectiveness of the prediction model, and re-establishing the baseline week during the sliding time window, that is, taking the baseline week of the sliding time window as the new prediction starting point. This approach of dynamically adjusting the prediction baseline can better reflect the real-time changes in the disease state, thereby improving the accuracy of the prediction.

[0101] In any of the above technical solutions, the HBsAg level decline rate data of the patient from week 0 to week 24 is used, and the 0 week at the time of initial diagnosis is taken as the baseline week. The time effect prediction model and the group trajectory model are coupled to construct the following time-plus-group dual-effect prediction model of the HBsAg level decline rate relative to the baseline based on the group trajectory model:

[0102]

[0103] in, It represents the relative decrease rate of HBsAg level of patient i at the tth detection time week after admission under the time plus group double effect prediction model; is the patient's baseline week; The number of weeks that the patient will participate in the forecast, that is, w i,t The most recent test week for the patient before the week; For this patient The actual decline rate of HBsAg relative to the baseline at each time week; is the number of weeks for the forecast interval; Indicates that the patient is in w i,t The intercept of the time effect when the weekly forecast is the jth latent class group, its value is shown in Table 2, j = 1, 2, 3; x 1,i Indicates gender, male is 1, female is 0; x 2,i represents the patient's age, x 3,i represents the patient's BMI value, represents the baseline HBsAg level of patient i, represents the baseline ALT level of patient i; x 6,i represents the baseline HBeAg level of patient i. If the baseline HBeAg test value is greater than 1, then x6,i =1, indicating HBeAg positive, otherwise x 6,i =0, indicating HBeAg negative; x 7,i represents the baseline HBVDNA level of patient i. If the baseline HBVDNA detection value is greater than the detection limit, then x 7,i =1, indicating viremia, otherwise x 7,i =0, indicating non-viremia.

[0104] Table 2. Intercepts of time plus group double effect at different prediction time intervals in the time plus group double effect prediction model

[0105]

[0106] By coupling the time effect prediction model with the group trajectory model, the time factor and the heterogeneity of the patient population are considered at the same time, so as to improve the accuracy of the prediction of the future HBsAg level decline rate. Using the HBsAg level decline rate data of patients from 0 to 24 weeks, the prediction can be dynamically updated according to the patient's latest test results, making the prediction results closer to the patient's actual condition. By introducing the concept of latent category groups, patient groups with similar disease decline trends can be identified, which helps doctors adopt more targeted treatment strategies for different patient groups.

[0107] In any of the above technical solutions, in step S5,

[0108] like Figure 2 As shown, when the number of weeks to be predicted does not exceed 24 weeks, the following steps are taken:

[0109] S51. Determine the baseline week and select the starting week of the initial diagnosis and treatment or study as the baseline week;

[0110] S52. Based on the baseline week data and the time effect prediction model, directly predict the decline rate of HBsAg level relative to the baseline in the next no more than 24 weeks.

[0111] By determining the baseline week and selecting the starting week of the initial diagnosis or study as the baseline, the initial state of disease development can be captured more accurately, providing a more reliable starting point for subsequent predictions. When the number of weeks to be predicted does not exceed 24 weeks, predictions are made directly based on the baseline week data and the time effect prediction model, avoiding complex data processing and model adjustments and improving the efficiency of predictions. This makes this method suitable for predicting the disease decline rate in the short term, which is of great significance for clinical decision-making or research that requires a quick understanding of the disease development trend.

[0112] In any of the above technical solutions, in step S5,

[0113] like Figure 3As shown, when the number of weeks to be predicted exceeds 24, the following steps are taken:

[0114] S53, using a time window sliding method to calibrate the time effect prediction model;

[0115] S54, setting the time window length. Except for the baseline week, the data time window length used in constructing the time effect prediction model is 24;

[0116] S55, when making a prediction, keep the time window length at 24 weeks, and slide the time window from the time of the first diagnosis in steps of 1 week. As time goes by, the time window will include the number of weeks to be predicted;

[0117] S56, re-establishing the baseline week, and taking the starting week of the sliding time window as the new baseline week after each sliding time window;

[0118] S57. Based on the data of the sliding time window and the established time effect prediction model, predict the future decline rate of HBsAg level relative to the baseline.

[0119] When the prediction period exceeds 24 weeks, the time effect prediction model is corrected by sliding the time window. This method allows the time effect prediction model to dynamically adjust according to the passage of time to better adapt to the needs of long-term prediction, thereby improving the accuracy and practicality of the prediction model. By sliding the time window, this method can include the number of weeks to be predicted over time, thereby realizing the prediction of long time periods, so that the method can adapt to the prediction needs of different lengths and has a wider range of applications. After each sliding time window, the baseline week is re-established, and the starting week of the sliding time window is used as the baseline week. The flexible baseline adjustment method enables the time effect prediction model to adapt to data changes in different time periods, further improving the accuracy of the prediction. This method is based on the data of the sliding time window and the established time effect prediction model to predict the future decline rate of HBsAg levels relative to the baseline. This prediction method based on real-time data can obtain prediction results more quickly and improve work efficiency.

[0120] In any of the above technical solutions, in step S1, the basic information includes gender, age and BMI, and the time series detection data includes detection date, HBsAg level, HBeAg level, ALT level and HBVDNA level.

[0121] By considering the patient's gender, age, BMI and other factors, it can provide each patient with a personalized disease decline rate prediction, which helps doctors develop more personalized treatment plans. By collecting time series detection data, it can monitor the patient's condition in real time, promptly detect signs of worsening or improvement, and improve prediction accuracy, providing doctors with timely feedback information to adjust treatment plans.

[0122] Example 2

[0123] Patient i, male (x 1,i =1), 50 years old (x 2,i =50) and BMI of 28.9 (x 3,i =28.9), baseline HBsAg level at first diagnosis Baseline ALT level The baseline HBeAg test value is 0.31<1, so x 6,i =0, indicating that the patient is HBeAg negative, and the baseline HBVDNA test value is 2570, which is greater than the detection limit of 100, so x 7,i =1, indicating viremia. It is known that the HBsAg values ​​after 1 week, 2 weeks and 3 weeks after the initial diagnosis were 733.58, 743.48 and 699.98 respectively. What is the relative decrease rate of HBsAg from the baseline in the 4th week and the 26th week after the initial diagnosis?

[0124] When predicting the HBsAg decline rate relative to baseline in the 4th week after initial diagnosis, w i,t =4<24, Using the data from week 3 for forecasting, To predict the time interval, based on the given information, Substituting into the time effect prediction model, the predicted result of HBsAg relative baseline decline rate at week 4 after initial diagnosis is:

[0125] When predicting the HBsAg decline rate relative to baseline at week 26 after initial diagnosis, because at this time w i,t =26>24, so the baseline week number cannot be directly taken Time window sliding is required. Based on the known information about the patient, Using the data from week 3 for forecasting, And there is a prediction time interval According to the information given, Assume that the patient's ALT level is 150 in week 2, that is Substituting into the time effect prediction model formula, the predicted result of HBsAg relative baseline decline rate at week 26 after initial diagnosis is:

[0126] Example 3

[0127] Patient i, female (x 1,i =0), 35 years old (x 2,i =35) and BMI of 23.5 (x 3,i =23.5), baseline HBsAg level at first diagnosis Baseline ALT level The baseline HBeAg test value is 0.4<1, so x 6,i =0, indicating that the patient is HBeAg negative and the baseline HBVDNA detection value is less than the detection limit of 100, so x 7,i =0, indicating that it is non-viremia. It is known that the HBsAg values ​​after 1 week, 3 weeks, 10 weeks, and 20 weeks after the initial diagnosis were 720.8, 680.7, 580.45, and 412.32, respectively. What is the relative decrease rate of HBsAg from the baseline at the 26th week after the initial diagnosis?

[0128] In order to predict the relative baseline decline rate of HBsAg at week 26 after initial diagnosis, we first calculated the conditional probabilities of the patient belonging to the three potential categories based on the known HBsAg test data before week 20 using the group trajectory model, which were 9.81%, 48.22%, and 41.97%, respectively. Therefore, we predicted that the patient belonged to the second trajectory group at week 26. Then, we used the time plus group double effect prediction model to predict the relative baseline decline rate of HBsAg. i,t =26>24, so the baseline week number cannot be directly taken Time window sliding is required. Based on the known information about the patient, Using the patient's most recent week 20 data from 26 weeks ago for prediction, And there is a prediction time interval At this time, the time plus group double effect prediction model Assume that the patient's ALT level is 50 at week 3, that is Then, by substituting the time plus group double effect prediction model, the prediction result of HBsAg relative baseline decline rate at week 26 after initial diagnosis can be obtained as follows:

[0129] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for predicting disease decline rate by mixed effect model, characterized in that: The following steps are involved: S1. Input the patient's basic information and time series test data; S2. Calculate the decline rate of the patient's HBsAg level relative to the baseline at each testing moment based on the patient's time series test data; S3. Establish a group trajectory model, use the conditional probability formula to calculate the probability that the patient belongs to different heterogeneous latent class groups, and determine the heterogeneous latent class group to which the patient is most likely to belong; S4. Establish a time effect prediction model and use the patient's HBsAg level decline rate data to predict the patient's future HBsAg level decline trend; S5. Couple the time effect prediction model and the group trajectory model to improve the prediction accuracy of the patient's future HBsAg level decline rate; In step S3, for patient i, the detection time point is w i,0 ,w i,1 ,w i,2 ,…,w i,t , where w i,0 represents the time of hospitalization and treatment, recorded as 0 week, w i,1 Indicates the number of weeks after the first test after hospitalization, w i,2 Indicates the number of weeks after hospitalization for the second test, w i,t represents the number of weeks after patient i was admitted to the hospital at the time of the tth test, w i,0 ,w i,1 ,w i,2 ,…,w i,t It is not necessarily a continuous positive integer; the actual decrease rate of HBsAg level relative to the baseline at each detection time week is Calculated by the following formula: in, represents the baseline HBsAg level, represents the HBsAg level at the tth detection time week, It represents the decline rate of HBsAg level of patient i at the tth detection time week relative to the baseline; according to The predicted decrease rate of HBsAg level relative to the baseline at each detection time week of the patient i when he belongs to different heterogeneous latent class groups is calculated and recorded as and Let T = count{w i,1 ,w i,2 ,…,w i,t }, represents the number of detection time points other than the baseline detection time, according to the conditional probability formula: The probability value of patient i belonging to different heterogeneous latent class groups is calculated, where σ1=0.17052475, σ2=0.28058448, σ3=0.30667541, y i,t represents the decline rate of HBsAg level relative to the baseline of patient i at the tth detection time, is the predicted decline rate of HBsAg level relative to the baseline of patient i belonging to trajectory group j at the tth detection week, Φ(x) is the standard normal cumulative distribution function, and its calculation formula is Compare the three conditional probability values ​​P(j|Y i ), j = 1, 2, 3, where patient i belongs to the heterogeneous latent class group corresponding to the maximum conditional probability value; In step S4, using the HBsAg level decline rate data of patient i from week 0 to week 24, with week 0 of the initial diagnosis as the baseline week, the time effect prediction model of the relative baseline decline rate of HBsAg level of the patient in different detection weeks is: in, It represents the relative decrease rate of HBsAg level of patient i at the tth detection time week after admission to hospital under the time effect prediction model; is the patient's baseline week; is the number of weeks that patient i participates in the prediction, that is, w i,t The most recent detection week of patient i before week; For patient i in The actual decline rate of HBsAg level relative to the baseline in the week; Indicates the forecast time interval in weeks; represents the time effect intercept at different prediction time intervals; x 1,i Indicates gender, male is 1, female is 0; x 2,i represents the patient's age, x 3,i represents the patient's BMI value, represents the baseline HBsAg level of patient i, represents the baseline ALT level of patient i; x 6,i represents the baseline HBeAg level of patient i. If the baseline HBeAg test value is greater than 1, then x 6,i =1, indicating HBeAg positive, otherwise x 6,i =0, indicating HBeAg negative; x 7,i represents the baseline HBVDNA level of patient i. If the baseline HBVDNA detection value is greater than the detection limit, then x 7,i =1, indicating viremia, otherwise x 7,i =0, indicating non-viremia; When the number of weeks to be predicted does not exceed 24 weeks, that is, w i,t ≤24, baseline weeks The time-effect prediction model was directly used to predict the decline rate of HBsAg level relative to the baseline; Time effect intercept The maximum number of weeks is only 24 weeks. When the number of weeks to be predicted exceeds 24 weeks, that is, w i,t >24, the time effect prediction model was corrected by sliding the time window; except for the baseline week, the time window length of the data used in the construction of the time effect prediction model was 24 weeks, and the time window length was set to 24 weeks; When the number of weeks to be predicted is w i,t >24, keep the time window length as 24 weeks and slide the time window from the time of the first diagnosis in steps of 1 week until the sliding time window just contains the week number w to be predicted i,t ; At this time, the baseline week is re-established for the sliding time window, that is, Week is the baseline week of the sliding time window; thus, the time effect prediction model is continued to be used to predict the wth week. i,t The rate of decrease of weekly HBsAg level relative to baseline; Using the data of HBsAg level decline rate from week 0 to week 24 of the patients, with week 0 of the initial diagnosis as the baseline week, the time effect prediction model and the group trajectory model were coupled to construct the following time plus group dual effect prediction model of HBsAg level decline rate relative to baseline based on the group trajectory model: in, It represents the relative decrease rate of HBsAg level of patient i at the tth detection time week after admission under the time plus group double effect prediction model; is the patient's baseline week; The number of weeks that the patient will participate in the forecast, that is, w i,t The most recent test week for the patient before the week; For this patient The actual decline rate of HBsAg relative to the baseline at each time week; is the number of weeks for the forecast interval; Indicates that the patient is in w i,t The time effect intercept when the weekly forecast is the jth latent class group, j = 1, 2, 3; x 1,i Indicates gender, male is 1, female is 0; x 2,i represents the patient's age, x 3,i represents the patient's BMI value, represents the baseline HBsAg level of patient i, represents the baseline ALT level of patient i; x 6,i represents the baseline HBeAg level of patient i. If the baseline HBeAg test value is greater than 1, then x 6,i =1, indicating HBeAg positive, otherwise x 6,i =0, indicating HBeAg negative; x 7,i represents the baseline HBVDNA level of patient i. If the baseline HBVDNA detection value is greater than the detection limit, then x 7,i =1, indicating viremia, otherwise x 7,i =0, indicating non-viremia; In step S5, when the number of weeks to be predicted does not exceed 24 weeks, the following steps are performed: S51. Determine the baseline week and select the starting week of the initial diagnosis and treatment or study as the baseline week; S52. Based on the baseline week data and the time effect prediction model, directly predict the decline rate of HBsAg level relative to the baseline in the next no more than 24 weeks.

2. The method for predicting disease decline rate by mixed effect model according to claim 1, characterized in that: In step S3, the heterogeneous latent class groups include a fast high clearance group, a delayed high clearance group and a slow low clearance group.

3. The method for predicting disease decline rate by mixed effect model according to claim 1, characterized in that: In the step S5, When the number of weeks to be forecasted exceeds 24, the following steps are taken: S53, using a time window sliding method to calibrate the time effect prediction model; S54, setting the time window length. Except for the baseline week, the data time window length used in constructing the time effect prediction model is 24; S55, when making a prediction, keep the time window length at 24 weeks, and slide the time window from the time of the first diagnosis in steps of 1 week. As time goes by, the time window will include the number of weeks to be predicted; S56, re-establishing the baseline week, and taking the starting week of the sliding time window as the new baseline week after each sliding time window; S57. Based on the data of the sliding time window and the established time-plus-group dual-effect prediction model, predict the future decline rate of HBsAg level relative to the baseline.

4. The method for predicting disease decline rate by mixed effect model according to claim 1, characterized in that: In step S1, the basic information includes gender, age and BMI, and the time series detection data includes detection date, HBsAg level, HBeAg level, ALT level and HBV DNA level.

Images