Method, system and device for evaluating HIV antiviral therapy compliance and storage medium

Abstract

The invention belongs to the field of medical research, and particularly relates to a method, system and device for evaluating HIV antiviral therapy compliance and a storage medium. The method comprises the following steps: acquiring index data for reflecting ART compliance of an HIV infected person, wherein the index data comprises a nursing retention state, a drug holding rate and a CD4 cell count; constructing a group trajectory model GBTM, and identifying trajectory subgroups of different antiviral treatment compliance in the HIV infected person based on the index data; utilizing a Logistic regression model to evaluate the influence of the demology and ART variable level of different HIV infected persons on the affiliation probability of each trajectory subgroup; utilizing a Cox proportional risk model to evaluate the association between each track subgroup and all-factor death, AIDS death and non-AIDS death; and finally, constructing a dynamic prediction model based on the nursing retention, the drug coverage rate and the CD4 cell counting data by using the combined model, so as to realize real-time early warning of the death risk.

Description

Technical Field

[0001] The present invention belongs to the field of medical research, and particularly relates to a method, system, device and storage medium for evaluating the compliance of HIV antiviral treatment. Background Art

[0002] The compliance of HIV antiviral treatment is the core link in the prevention and treatment of AIDS, which directly determines the virus suppression effect, the quality of life of patients and the risk of disease transmission. If there is poor compliance for a long time, the virus is likely to produce drug-resistant mutations, which not only leads to treatment failure, but also accelerates the disease progression, increases the all-cause mortality rate, and at the same time increases the risk of community transmission. Important progress has been made in the research on the association between the antiviral treatment compliance of global HIV-infected patients and clinical outcomes, but there are still key limitations such as insufficient dynamic trajectory analysis, lack of longitudinal prediction models and lack of multi-dimensional mechanism analysis.

[0003] Existing related technologies such as Chinese Patent Document CN114708989A disclose a detection method for the pharmaceutical service effect based on AIDS patients. Through the experimental detection of the viral load level of HIV-infected patients and the analysis of the results of the peripheral blood CD4 + T lymphocyte count and the gene detection results of HIV drug-resistant sites of AIDS patients, the viral load levels of two groups of patients who received pharmaceutical services and those who did not receive pharmaceutical services are compared, and the changes in the CD4 + T lymphocyte count, and the results of the drug-resistant site situation are generated; the invention hopes to prove the correlation between clinical pharmaceutical service treatment and the medication compliance of AIDS patients based on experimental data through relevant investigation and research, and hopes to provide evidence support for carrying out clinical pharmaceutical services for AIDS patients on a larger scale, and explore a specialized and standardized pharmaceutical service model process for AIDS patients.

[0004] In addition, Chinese Patent Document CN119724575A discloses a computer device and a computer-readable storage medium for evaluating the immune reconstruction status after ART of HIV-infected patients, and the technical problem to be solved is how to evaluate the immune reconstruction status after ART of HIV-infected patients. The computer program stored in the memory in the computer device provided by the invention is: S1. Receive sample data, which is the proportion of the number of PD-1-expressing cells in CD8<supgt;+< / supgt>T cells of peripheral blood mononuclear cells of HIV-infected subjects after ART and the proportion of the number of cells expressing at least one of the following 5 factors: CD38, HLA-DR, CD57, PERFORIN, and CD107A; S2. Obtain a predicted value Y from the sample data through a model formula, and output the immune reconstruction situation according to the predicted value Y and the pre-determination condition; the invention can accurately and efficiently determine whether the immune reconstruction status of HIV-infected patients after antiviral treatment is good.

[0005] In studies on HIV antiretroviral therapy adherence, most literature, when exploring the association between adherence and mortality, has not fully considered the longitudinal dynamic changes in adherence, focusing only on adherence at a few time points, making it difficult to accurately characterize the fluctuations in patients' adherence throughout the entire treatment cycle. Previous studies have mostly been based on single-time-point adherence indicators, failing to dynamically track the trajectory of adherence changes and individual differences, making it difficult to update mortality risk predictions in real time. Furthermore, there is insufficient analysis of multidimensional factors affecting antiretroviral therapy adherence, such as a lack of comprehensive consideration of the interaction between patients' psychosocial factors and clinical factors (treatment regimen, underlying diseases).

[0006] While joint models (such as the Jointmodels-Bayes framework), mixed-effects models combined with Cox models offer theoretical possibilities for dynamically predicting the survival risk of HIV-infected individuals, current domestic and international literature on the dynamic prediction of outcomes such as mortality based on treatment adherence is still very scarce. Existing studies are mostly based on cross-sectional surveys or short-term cohort analyses, making it difficult to capture the dynamic changes in treatment adherence over time and its long-term impact on mortality risk.

[0007] Based on this, the present invention designs a method for evaluating HIV antiviral treatment adherence, aiming to solve the problems of insufficient dynamic trajectory analysis, lack of longitudinal prediction, and lack of multi-dimensional mechanism analysis in existing methods. Summary of the Invention

[0008] The present invention aims to overcome at least one of the defects of the prior art and provide a method for evaluating HIV antiviral treatment adherence.

[0009] The detailed technical solution of this invention is as follows: A method for assessing HIV antiretroviral therapy adherence, the method comprising: S1. Obtain indicator data to reflect ART adherence in HIV-infected individuals, including care retention status, medication possession rate, and CD4 cell count. S2. Construct the group trajectory model GBTM, and based on the index data, identify different trajectory subgroups of antiretroviral therapy adherence among HIV-infected individuals. S3. Use a logistic regression model to calculate the odds ratio (OR) and 95% confidence interval (CI) for each trajectory subgroup to assess the impact of different HIV-infected individuals’ demographic and ART variable levels on the probability of belonging to each trajectory subgroup. S4. Calculate the hazard ratio (HR) and 95% confidence interval (CI) for each trajectory subgroup using the Cox proportional hazards model to assess the association between each trajectory subgroup and all-cause mortality, AIDS mortality, and non-AIDS mortality, respectively. S5. Using the random effects model framework in the joint model, the index data are used as longitudinal endogenous variables to construct a predictive model for the prognosis of antiretroviral therapy for HIV-infected individuals, so as to dynamically predict the risk of future death.

[0010] According to a preferred embodiment of the present invention, in S2, the basic form of the group trajectory model GBTM is:

[0011] In formula (1): Represents a random vector The probability of a random vector Defined as an individual exist Measurement values ​​of variables at each time point The vertical sequence; The group trajectory model GBTM represents the random vectors of individuals. The changing trend of the number of trajectory subgroups in the cluster; For individuals Belongs to the The probability of a group is expressed as the probability of randomly selecting an individual from the population. Belongs to the The probability of the group; Represents the given first Group of random vectors The conditional probability distribution function.

[0012] According to a preferred embodiment of the present invention, in S2, the individual Belongs to the group probability The calculation is as follows:

[0013] In formula (2): Indicates the first Group weights.

[0014] According to a preferred embodiment of the present invention, in S2, the first Group of random vectors conditional probability distribution function The calculation is as follows:

[0015] In formula (3): Represents an individual Belongs to the The group is at the point in time. Observations The probability distribution function; Represents an individual At the point of time The observed values, namely ART adherence, are reflected by care retention status, medication holding rate, and CD4 cell count; in, This polynomial function is expressed as:

[0016] In equation (4): Representing continuous time points and observations Latent variables; Represents an individual At the point of time The random error term follows a normal distribution. That is, the mean is 0 and the variance is ; The intercept determines the starting point height; The coefficient of the linear term determines the overall upward or downward trend; The coefficient of the quadratic term determines the direction of curvature of the curve, i.e., whether it is convex or concave. These are the coefficients of the cubic term, which determine the complex curvature of the curve, such as the S-shape; each trajectory is formed by the combined effect of these coefficients to create a trajectory of a specific shape. Represents an individual At the point of time The follow-up time is the time represented by the horizontal axis of the trajectory; Based on the range of values ​​for the indicator data, the maximum value is set to be... The minimum value is ;like < ,but = ;like > ,but = ;like < < ,but = .

[0017] According to a preferred embodiment of the present invention, in S4, a Cox proportional hazards model is used to assess the association between each trajectory subgroup and the risk of mortality. Based on the different independent variables included in the Cox proportional hazards model, the following three models are defined to control for confounding effects: The first model is without covariate adjustment, that is, it does not include covariates and only inputs trajectory subgroups; The second model adjusts for demographic variables including age, gender, ethnicity, place of residence, education level, marital status, and body mass index (BMI). The third model is based on the second model and further adjusts the following variables: initial CD4 cell count, ART regimen, whether the ART regimen has been changed, route of infection, drug side effects, whether the patient has received sulfamethoxazole-trimethoprim for the prevention of opportunistic infections, and clinical and treatment variables related to sexually transmitted diseases.

[0018] According to a preferred embodiment of the present invention, in S5, the random effects model framework in the joint model includes a longitudinal sub-model, which is used to capture the dynamic changes of the indicator data to output individual-level trend parameters, wherein: A generalized linear mixed-effects model was used to fit the changes in retention probability over follow-up time for nursing care retention status. A linear mixed-effects model was used to fit the numerical fluctuation trends of drug holding rate and CD4 cell count over time.

[0019] According to a preferred embodiment of the present invention, in S5, the random effects model framework in the joint model further includes a survival sub-model, which uses a Cox proportional hazards model to correlate the individual-level trend parameters output by the longitudinal sub-model with the risk of death and outputs a risk prediction formula to dynamically predict the risk of future death.

[0020] In another aspect of the invention, a system for assessing HIV antiretroviral therapy adherence is provided, employing the method for assessing HIV antiretroviral therapy adherence as described above, the system comprising: The data acquisition module is used to acquire indicator data reflecting the ART adherence of HIV-infected individuals, including care retention status, medication possession rate, and CD4 cell count. The trajectory grouping module is used to construct the group trajectory model GBTM and, based on the index data, identify trajectory subgroups with different antiretroviral therapy adherence among HIV-infected individuals. The first assessment module is used to calculate the odds ratio (OR) and 95% confidence interval (CI) for each trajectory subgroup using a logistic regression model, in order to assess the impact of different levels of HIV-infected individuals’ demographic and ART variables on the probability of belonging to each trajectory subgroup. The second assessment module is used to calculate the hazard ratio (HR) and 95% confidence interval (CI) for each trajectory subgroup using the Cox proportional hazards model, in order to assess the association between each trajectory subgroup and all-cause mortality, AIDS mortality, and non-AIDS mortality, respectively. The risk prediction module is used to construct a predictive model for the prognosis of antiretroviral therapy for HIV-infected individuals by using the random effects model framework in the joint model as longitudinal endogenous variables, so as to dynamically predict the risk of future death.

[0021] In another aspect of the invention, an apparatus is also provided, comprising: At least one processor; and A memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the method described above for assessing HIV antiretroviral therapy adherence.

[0022] In another aspect of the invention, a machine-readable storage medium is also provided, which stores executable instructions that, when executed, cause the machine to perform the method described above for assessing HIV antiretroviral therapy adherence.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention provides a method for assessing HIV antiretroviral therapy adherence. It adopts a progressive research framework of “epidemiological characterization - trajectory model construction - two-stage dynamic prediction - multi-dimensional mechanism analysis”. By integrating logit regression model and GBTM trajectory model and combined survival analysis method, it explores the association between HIV-infected individuals’ adherence trajectory type and mortality risk. It also uses structural equation modeling to analyze the action path of multi-dimensional influencing factors and creates a longitudinal prediction model for HIV antiretroviral therapy adherence.

[0024] (2) This invention takes into account the longitudinal changes in nursing care retention, predicts the mortality risk of HIV-infected individuals in real time, and fully considers the impact of multidimensional factors on HIV antiretroviral therapy adherence; and through this invention, the change patterns of adherence in the longitudinal data of HIV-infected individuals can be explored, high-risk situations such as high mortality risk and virological / immunological failure risk can be identified, providing targeted solutions for improving the prognosis of HIV-infected individuals and providing theoretical support for regional HIV research. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method for assessing HIV antiviral treatment adherence as described in this invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0029] In the case of no conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0030] Embodiment 1 Refer Figure 1 , this embodiment provides a method for evaluating the adherence to HIV antiviral therapy, and the method includes: S1. Obtain index data for reflecting the ART adherence of HIV-infected individuals, and the index data includes the nursing retention status, the medication possession ratio, and the CD4 cell count.

[0031] Specifically, in this embodiment, based on the cohort of HIV-infected individuals receiving HIV antiviral therapy in Shandong Province from 2000 to 2024, through spatio-temporal-population distribution analysis, the baseline characteristics and dynamic changes of the research subjects are characterized. The dimensions of characterization include: time, space, human characteristics, and the trend of change over time; the variables of characterization include: (1) general demographic characteristics, including gender, age, marital status, educational level, etc.; (2) follow-up treatment information, including the start time of treatment, the CD4 detection time, the viral load detection time, the medication collection time, the follow-up date, the follow-up content, the treatment regimen, etc.; (3) death information, including the death time, the cause of death, etc.

[0032] Based on the above data, index data for reflecting the ART adherence of HIV-infected individuals is further obtained, including the nursing retention status, the medication possession ratio, and the CD4 cell count.

[0033] The nursing retention status described in this embodiment is defined as: in each natural year after the start of antiviral therapy, if the HIV-infected individual completes ≥2 CD4 or viral load detections, and the interval between at least 2 detections > 3 months, it is defined as the "nursing retention status" (good adherence); otherwise, it is the "non-nursing retention status" (poor adherence).

[0034] According to the antiretroviral therapy manual, all HIV-infected individuals must have regular follow-up visits to the treatment center after starting treatment. Follow-up visits are required at 0.5, 1, 2, and 3 months, and then every 3 months thereafter. During these visits, doctors provide medical consultations, patients receive their antiretroviral medications, and necessary laboratory tests are conducted. Subsequently, healthcare staff will record each follow-up visit information in the antiretroviral therapy follow-up database, including the visit date, content of the visit, treatment plan, and expected next medication refill date.

[0035] Medication Possession Ratio (MPR) refers to the ratio of the number of days an HIV-infected person actually possesses medication within a specific period after their first dose to the total number of days in that period. It is an important indicator for measuring medication use behavior in HIV-infected individuals. Calculating MPR through detailed analysis of HIV-infected individuals' electronic prescription records indirectly reflects medication adherence. This method, which involves secondary processing of the database to quantify and accurately assess medication adherence, is currently the most internationally accepted method for this purpose. Compared to other methods of assessing adherence, MPR provides a more objective assessment with the lowest risk of overestimation or underestimation. The formula for calculating MPR is: MPR = (Number of days with medication coverage during follow-up / Total number of follow-up days) * 100%.

[0036] CD4 cell count refers to the actual number of CD4 tests performed divided by the theoretical number of CD4 tests that should be performed during the study period according to national guidelines. CD4 tests are performed in the 3rd, 6th, and 12th months of the first year after ART begins, and then every 6 months thereafter.

[0037] This embodiment implements the above steps based on the following data packet: The 'tidyverse' package (which includes 'dplyr's' 'filter()' for data filtering, 'group_by()' + 'summarise()' for grouping and calculating statistics, and 'tidyr's' 'pivot_longer()' / 'pivot_wider()' for data format conversion), the 'summarytools' package's 'descr()' function for generating structured baseline feature description tables, and the 'ggplot2' package's 'ggplot()' + 'geom_bar()' (for categorical variable bar plots) / 'geom_boxplot()' (for continuous variable box plots) for visualization.

[0038] S2. Construct the group trajectory model GBTM, and based on the index data, identify different trajectory subgroups of antiretroviral therapy adherence among HIV-infected individuals.

[0039] In this embodiment, considering the dynamic nature of ART adherence, this method aims to explore the relationship between ART adherence (including care retention, medication holding rate, and CD4 cell count) and its dynamic changes. ART adherence is influenced not only by individual factors of HIV-infected individuals but may also be affected by the interaction of various external factors during treatment. Therefore, in-depth analysis of the trajectory of ART adherence changes is of great significance for revealing its patterns and its impact on the health outcomes of HIV-infected individuals.

[0040] To further capture the potential heterogeneity of adherence trajectories, this embodiment employs group-based trajectory modeling (GBTM) to identify different trajectory subgroups of antiretroviral therapy adherence among HIV-infected individuals based on index data reflecting ART adherence.

[0041] Specifically, the GBTM model will integrate individual... exist Measurement values ​​of variables at each time point The vertical sequence is defined as a random vector. Here, the individual This refers to individuals infected with HIV. The model assumes that the entire population can be determined based on the random vectors of individuals. Clustering of changing trends There are three heterogeneous subgroups, where individuals within each subgroup have similar developmental trajectories, while the trajectories between different subgroups show significant differences. Let... Represents a random vector The probability is used to obtain the basic form of the GBTM model:

[0042] In formula (1): For individuals Belongs to the The probability of a group is expressed as the probability of randomly selecting an individual from the population. Belongs to the The probability of the group; Represents the given first Group of random vectors The conditional probability distribution function.

[0043] Among them, individuals Belongs to the group probability The calculation is as follows:

[0044] In formula (2): Indicates the first Group weights.

[0045] Represents the given first Group of random vectors The conditional probability distribution function. To simplify model construction, it is usually assumed that the variables have conditional independence across multiple observations. That is, it is assumed that... Within the given time point, for the given... Groups, random vectors Observations They are conditionally independent, therefore It can be represented as:

[0046] In formula (3): Represents an individual Belongs to the The group is at the point in time. Observations The probability distribution function; Represents an individual At the point of time The observed value, namely ART adherence, is reflected by care retention status, medication holding rate, and CD4 cell count.

[0047] Based on the data type and structure of the research materials, Different polynomial function equations can be used for fitting. Currently, the model can adapt to the following distribution forms: censored normal distribution (CNORM), Poisson distribution, logit distribution, and beta distribution. In the CNORM distribution form, i.e., the data type satisfied in this embodiment, This polynomial function can be expressed as:

[0048] In equation (4): Representing continuous time points and observations Latent variables; Represents an individual At the point of time The random error term follows a normal distribution. That is, the mean is 0 and the variance is ; The intercept determines the starting point height; The coefficient of the linear term determines the overall upward or downward trend; The coefficient of the quadratic term determines the direction of curvature of the curve, i.e., whether it is convex or concave. These are the coefficients of the cubic term, which determine the complex curvature of the curve, such as the S-shape; each trajectory is formed by the combined effect of these coefficients to create a trajectory of a specific shape. Represents an individual At the point of time The follow-up time is the time represented by the horizontal axis of the trajectory.

[0049] Based on the range of nursing retention status, medication holding rate, and CD4 cell count, the maximum value is set to be... The minimum value is ;like < ,but = ;like > ,but = ;like < < ,but = .

[0050] Although the trajectory model is conceptually based on the aim of identifying subgroups with similar trajectories, the parameter estimates of the model are not the result of cluster analysis, but rather derived from maximum likelihood estimation, which is achieved through the maximum likelihood method.

[0051] Therefore, according to the above formulas (1) to (3), the GBTM model can assume that it is in the first position. Calculate individuals under group conditions Measured values probability distribution function This allows us to determine the trajectory of each individual and assign each individual to the trajectory subgroup corresponding to the highest probability.

[0052] Since the GBTM model is an unsupervised clustering model, determining the optimal model requires specific selection criteria. The selection process typically involves two steps: First, the highest polynomial degree of the trajectory is fixed at cubic, and the number of subgroups is gradually increased from one, usually up to a maximum of five. Selection is based on the lowest Bayesian information criterion (BIC) value, an average posterior probability of assignment > 0.70, and a group membership ratio ≥ 5.0%. Second, after determining the number of trajectory subgroups, trajectories are fitted starting with higher-order cubic terms. If higher-order polynomials are not statistically significant, the polynomial degree is reduced. It is important to note that even if a linear term is not statistically significant, it will still be retained in the final model.

[0053] In this embodiment, although age and survey period can both be used as time points. However, since the data mainly came from electronic medical records and referenced previous research, the survey period was chosen as the time point, with each calendar year defined as a survey period. The study explored models with a maximum of four trajectory subgroups and a maximum of three cubic terms, and determined the optimal model based on the model selection criteria. The optimal model was ultimately determined to be a three cubic term model with five trajectory subgroups. .

[0054] Finally, the optimal GBTM model obtained above was used to identify heterogeneous trajectory subgroups of HIV-infected individuals in terms of care retention, medication holding rate, and CD4 cell count.

[0055] The specific implementation is as follows: the base trajectory model is constructed using the 'traj()' function of the 'traj' package (defining trajectory variables, fitting a multinomial model, and specifying the distribution type to divide the trajectory subgroups), and SAS version 9.4 is only used to run the built-in analysis program of the trajectory model.

[0056] Based on the above, the final trajectory subgroups include the low compliance group, the compliance U-shaped group, and the high compliance declining group.

[0057] In the population trajectory analysis, the "compliance U-shaped group" refers to those whose measured variables (in this example, nursing retention status, medication holding rate, and CD4 cell count) showed a trend of "deteriorating first and then improving." Specifically, this group was at a relatively good level at the beginning of the study, but gradually deteriorated to a low point over time, and then showed a significant rebound and improvement in the later stages of the study, with their overall change path resembling a "U" shape.

[0058] It should be understood that the trajectory subgroups of a variable do not necessarily have to fall into these three categories; they may also include stable groups, rising groups, higher levels, etc. Analysis based on data from Shandong Province reveals the aforementioned three trajectory subgroups.

[0059] S3. Calculate the odds ratio (OR) and 95% confidence interval (CI) for each trajectory subgroup using a logistic regression model to assess the impact of different HIV-infected individuals’ demographic and ART levels on the probability of belonging to each trajectory subgroup.

[0060] This embodiment aims to further explore the association between demographic and ART-related variables of HIV-infected individuals and ART adherence trajectories, using a population-based GBTM model and a multivariate logistic regression model for analysis.

[0061] Specifically, the study subjects were first divided into different trajectory subgroups using the GBTM model. Then, the trajectory subgroups were used as the dependent variable, and HIV-infected individuals' demographic and ART-related variables were used as independent variables. ART-related variables mainly included the start time of treatment, treatment regimen, baseline CD4 cell count, viral load, drug side effects, whether the treatment regimen was changed, and whether the individual received sulfamethoxazole-trimethoprim for opportunistic infection prevention. Next, a multinomial logistic regression model was used to analyze the impact of demographic, clinical, and treatment-related factors on the subgroup assignment. Specifically, the 'multinom()' and 'predict()' functions of the 'nnet' package were used to calculate the odds ratio (OR) and 95% confidence interval (CI) of each trajectory subgroup relative to the control group to assess the impact of different levels of HIV-infected individuals' demographic and ART-related variables on the probability of assignment to each trajectory subgroup.

[0062] S4. Calculate the hazard ratio (HR) and 95% confidence interval (CI) for each trajectory subgroup using the Cox proportional hazards model to assess the association between each trajectory subgroup and all-cause mortality, AIDS mortality, and non-AIDS mortality.

[0063] This embodiment utilizes a Cox proportional hazards model (with multi-stage covariate adjustment) to further explore the association between HIV-infected individuals' care retention trajectory and mortality risk. The analysis employs a population-based GBTM model and a Cox proportional hazards model. First, the GBTM model is used to divide the study subjects into different trajectory subgroups, and then the Cox proportional hazards model is used to explore the association between HIV-infected individuals' care retention trajectory and mortality risk.

[0064] Specifically, the survival differences among the trajectory subgroups can be verified by constructing a multi-stage covariate-adjusted Cox model using the 'coxph()' function of the 'survival' package to calculate the hazard ratio (HR) and 95% confidence interval (CI); and by using Kaplan-Meier survival analysis combined with the Log-rank test (i.e., estimating the survival curve using the 'survfit()' function of the 'survival' package, performing the test using the 'survdiff()' function, and plotting the survival curve with the hazard table using the 'ggsurvplot()' function of the 'survminer' package).

[0065] Specifically, the Cox proportional hazards model in this embodiment includes multi-stage covariate adjustment to more accurately explore the association between different trajectory subgroups and the risk of all-cause mortality, AIDS-related mortality, and non-AIDS-related mortality. Specifically, it controls for confounding effects through a model with three progressively adjusted covariates, defined as follows: The first model is without covariate adjustment, that is, it does not include covariates and only inputs trajectory subgroups; The second model adjusts for demographic variables including age, gender, ethnicity, place of residence, education level, marital status, and body mass index (BMI). The third model is based on the second model and further adjusts the following variables: initial CD4 cell count, ART regimen, whether the ART regimen has been changed, route of infection, drug side effects, whether the patient has received sulfamethoxazole-trimethoprim for the prevention of opportunistic infections, and clinical and treatment variables related to sexually transmitted diseases.

[0066] Using the low compliance group as a reference, the specific results are as follows: Table 1. Cox proportional hazards regression results of HIV-infected individuals' care retention trajectory subgroups and all-cause mortality.

[0067] As shown in Table 1, without adjusting for covariates, the HR for the compliance U-shaped group was 0.66 (95% CI: 0.53–0.82, P<0.001), and the HR for the high compliance decline group was 0.58 (95% CI: 0.45–0.75, P<0.001). After adjusting for demographic variables, the HRs for the two groups decreased to 0.64 (95% CI: 0.51–0.80, P<0.001) and 0.55, respectively, in the second model. (95% CI: 0.42–0.72, P<0.001); After the third model further incorporated clinical treatment variables, the HRs for the two groups continued to decrease to 0.59 (95% CI: 0.43–0.82, P=0.002) and 0.42 (95% CI: 0.29–0.62, P<0.001), suggesting that the risk of all-cause mortality was more significantly reduced in the high adherence decline group, and this association remained stable after controlling for multiple confounding factors.

[0068] Table 2. Cox proportional hazards regression results of HIV-infected individuals' care retention trajectory subgroups and AIDS-related deaths.

[0069] Table 2 shows the differences in the association between the two groups and the risk of AIDS-related death in different models. In the first model, without adjusting for covariates, the association in the adherence U-shaped group was nearly significant (HR=0.70, 95%CI: 0.49–1.02, P=0.062), while the high adherence-decreasing group showed a significant reduction in risk (HR=0.60, 95%CI: 0.39–0.93, P=0.021). In the second model, after adjusting for demographic variables, the adherence U-shaped group association became significant (HR=0.66, 95%CI: 0.45–0.95, P=0.027), and the HR in the high adherence-decreasing group decreased to 0.53 (95%CI: 0.3). 4~0.84, P=0.006); however, after further adjusting for clinical treatment variables in the third model, the association in the adherence U-shaped group became insignificant again (HR=0.61, 95%CI: 0.35~1.05, P=0.072), while the risk reduction in the high adherence decline group further increased (HR=0.35, 95%CI: 0.18~0.67, P=0.001), suggesting that the protective effect of the high adherence decline group against AIDS-related mortality is independent of confounding factors such as clinical treatment, while the association in the adherence U-shaped group may be affected by clinical factors.

[0070] Table 3. Cox proportional hazards regression results of HIV-infected individuals' care retention trajectory subgroups and non-AIDS-related deaths.

[0071] In Table 3, the negative associations between the adherence U-shaped group and the high adherence decline group and the risk of non-AIDS-related mortality remained stable and significant in all models, with the protective effect of the high adherence decline group consistently being stronger. In the first model, the HR for the adherence U-shaped group was 0.65 (95% CI: 0.48–0.87, P = 0.004), and the HR for the high adherence decline group was 0.54 (95% CI: 0.39–0.77, P = 0.001). In the second model, after adjusting for demographic variables, the HRs for the two groups slightly decreased to 0.63 (95% CI: 0.47–0.86, P = 0.003) and 0.53 (95% CI: 0.37–0.7) respectively. 5, P<0.001); After incorporating clinical treatment variables into the third model, the HRs for the two groups were 0.58 (95% CI: 0.38–0.91, P=0.018) and 0.47 (95% CI: 0.28–0.79, P=0.004), respectively, indicating that regardless of whether demographic and clinical treatment factors were adjusted, both non-low adherence trajectories significantly reduced the risk of non-AIDS-related death, and the protective effect was more prominent in the high adherence decline group.

[0072] In this embodiment, the follow-up period was from the baseline assessment date to death or the cutoff date (December 31, 2024), with death as the follow-up endpoint. The proportional hazards hypothesis was validated by the Schoenfeld residual test (P > 0.05). All analyses were adjusted for covariates such as age, sex, ethnicity, education level, Townsend deprivation index, and alcohol intake.

[0073] S5. Using the random effects model framework in the joint model, the index data are used as longitudinal endogenous variables to construct a predictive model for the prognosis of antiretroviral therapy for HIV-infected individuals, so as to dynamically predict the risk of future death.

[0074] This involves using a combined model to construct dynamic prediction models based on nursing retention status, medication holding rate, and CD4 cell count data, specifically through the synergistic interaction of longitudinal sub-models and survival sub-models. The longitudinal sub-model is used to capture dynamic changes in indicators and output individual horizontal trend parameters.

[0075] The core task of the longitudinal sub-model is to process indicators that are repeatedly measured over the follow-up period, such as nursing retention status (binary data, such as "retention = 1 / non-retention = 0"), drug holding rate (continuous data, such as 0-100%), and CD4 cell count (continuous data, such as cells / μL), to characterize the changes in indicators for each individual at different time points.

[0076] Specifically, the mixed-effects model adapted to the data type in the 'JMbayes2' package is adopted: A generalized linear mixed-effects model, such as the glmer function and the logit link function, was used to fit the change in retention probability over follow-up time for nursing care retention status. A linear mixed-effects model, such as the LME function, was used to fit the numerical fluctuation trend of drug holding rate and CD4 cell count over time.

[0077] Ultimately, the system will output "longitudinal dynamic parameters" for each individual, including: ① fixed effects (e.g., an average annual increase of 30 CD4 cell counts / μL at the population level); ② individual random effects (e.g., a patient's CD4 growth rate is 15 cells / μL faster than the population average, i.e., the individual-specific slope; a patient's baseline drug holding rate is 12% higher than the population average, i.e., the individual-specific intercept); ③ "predicted values" of the indicators at any follow-up time point (e.g., predicting a patient's CD4 level one year after treatment based on the patient's first three CD4 measurements). These results provide "dynamic indicator evidence" for subsequent associated mortality risk.

[0078] The survival sub-model uses the Cox proportional hazards model to correlate the individual-level trend parameters output by the longitudinal sub-model with mortality risk and outputs a risk prediction formula to dynamically predict future mortality risk.

[0079] The core task of the survival sub-model is to correlate the "dynamic indicator parameters" obtained from the longitudinal sub-model with mortality outcomes (all-cause mortality, AIDS-related mortality) to construct a risk prediction framework with time-dependent covariates. The 'jointModelBayes()' function from the 'JMbayes2' package is used to incorporate the "individual-specific indicator trends (such as CD4 growth slope, drug holding rate fluctuations)" and "predicted indicator values ​​at any time point (such as nursing care retention status at 6 months of treatment)" output by the longitudinal sub-model as time-dependent covariates into the Cox proportional hazards model.

[0080] The final output will be: ① the association strength between each dynamic indicator and mortality risk (e.g., for every 100 CD4 cell count increase / μL, the mortality risk decreases by 35%, HR=0.65, 95%CI 0.52-0.81; for every 10% decrease in drug possession rate, the mortality risk increases by 20%, HR=1.20, 95%CI 1.08-1.34); ② a "mortality risk prediction formula" based on dynamic indicators, such as the mortality risk of an individual at time t = baseline risk × exp[CD4 level × coefficient 1 + drug possession rate × coefficient 2 + nursing retention status × coefficient 3]; these results provide "quantitative association rules" for calculating real-time risk.

[0081] Dynamic prediction is achieved through the linkage of the two models described above: real-time data updates → iterative risk calculation.

[0082] The core logic of dynamic prediction is "continuously updating data and iteratively optimizing predictions as the follow-up process progresses": When an individual completes a follow-up (e.g., after 1 year of treatment), the new data on nursing retention status, medication holding rate, and CD4 cell count from that follow-up are first substituted into the longitudinal sub-model to update the individual's "dynamic parameters of the indicators" (e.g., if the original prediction was that the patient's CD4 count would increase by 30 cells / μL per year, it is adjusted to 38 cells / μL per year based on the newly measured CD4 value, and the accurate CD4 value after 1 year of treatment is obtained); then, the updated "dynamic parameters of the indicators" and "current indicator values" are substituted into the survival sub-model to calculate the individual's probability of death and 95% CI in a specific period after the current time point (e.g., the next 6 months or 1 year) (e.g., the predicted mortality risk in the next 6 months after 1 year of treatment is 2.8%, with a 95% CI of 1.5%-4.6%); when the individual completes the next follow-up (e.g., after 1.5 years of treatment), the above steps are repeated—the parameters of the longitudinal sub-model are updated with the new follow-up data, and then substituted into the survival sub-model to obtain a new risk prediction value. Through this iterative process of "following up once, updating once, and predicting once," "real-time early warning" of mortality risk changes with individual indicators is achieved, rather than relying solely on static predictions based on baseline data.

[0083] As described above, the joint model construction utilizes the 'lme()' / 'glmer()' functions from the 'JMbayes2' package to fit the longitudinal sub-model, the 'jointModelBayes()' function to associate the longitudinal sub-model with the survival sub-model, and the 'predict()' function to generate dynamic mortality risk predictions. In this embodiment, all statistical analyses, except for the trajectory model which was performed using SAS version 9.4, were completed using R version 4.4.3 software.

[0084] This invention allows us to explore patterns of compliance changes in longitudinal data of HIV-infected individuals, identify high-risk situations such as high risk of death and risk of virological / immunological failure, provide targeted solutions for improving the prognosis of HIV-infected individuals, and provide theoretical support for regional HIV research.

[0085] Example 2 This embodiment provides a system for assessing HIV antiretroviral therapy adherence, using the method described above for assessing HIV antiretroviral therapy adherence. The system includes: The data acquisition module is used to acquire indicator data reflecting the ART adherence of HIV-infected individuals, including care retention status, medication possession rate, and CD4 cell count. The trajectory grouping module is used to construct the group trajectory model GBTM and, based on the index data, identify trajectory subgroups with different antiretroviral therapy adherence among HIV-infected individuals. The first assessment module is used to calculate the odds ratio (OR) and 95% confidence interval (CI) for each trajectory subgroup using a logistic regression model, in order to assess the impact of different levels of HIV-infected individuals’ demographic and ART variables on the probability of belonging to each trajectory subgroup. The second assessment module is used to calculate the hazard ratio (HR) and 95% confidence interval (CI) for each trajectory subgroup using the Cox proportional hazards model, in order to assess the association between each trajectory subgroup and all-cause mortality, AIDS mortality, and non-AIDS mortality, respectively. The risk prediction module is used to construct a predictive model for the prognosis of antiretroviral therapy for HIV-infected individuals by using the random effects model framework in the joint model as longitudinal endogenous variables, so as to dynamically predict the risk of future death.

[0086] Example 3 This embodiment also provides a device, including: At least one processor; and A memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the method described above for assessing HIV antiretroviral therapy adherence.

[0087] In this embodiment, the device may include, but is not limited to: personal computer, server computer, workstation, desktop computer, laptop computer, notebook computer, mobile computing device, smartphone, tablet computer, cellular phone, personal digital assistant (PDA), handheld device, messaging device, wearable computing device, consumer electronic device, etc.

[0088] Example 4 This embodiment also provides a machine-readable storage medium storing executable instructions that, when executed, cause the machine to perform the method described above for assessing HIV antiretroviral therapy adherence.

[0089] Specifically, a system or apparatus equipped with a readable storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer or processor of the system or apparatus can read and execute the instructions stored in the readable storage medium.

[0090] In this case, the program code read from the readable medium itself can perform the functions of any of the above embodiments, and therefore the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of this specification.

[0091] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer or the cloud via a communication network.

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0093] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solutions of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for assessing HIV antiretroviral therapy adherence, characterized in that, The method includes: S1. Obtain indicator data to reflect ART adherence in HIV-infected individuals, including care retention status, medication possession rate, and CD4 cell count. S2. Construct the group trajectory model GBTM, and based on the index data, identify different trajectory subgroups of antiretroviral therapy adherence among HIV-infected individuals. S3. Use a logistic regression model to calculate the odds ratio (OR) and 95% confidence interval (CI) for each trajectory subgroup to assess the impact of different HIV-infected individuals’ demographic and ART variable levels on the probability of belonging to each trajectory subgroup. S4. Calculate the hazard ratio (HR) and 95% confidence interval (CI) for each trajectory subgroup using the Cox proportional hazards model to assess the association between each trajectory subgroup and all-cause mortality, AIDS mortality, and non-AIDS mortality, respectively. S5. Using the random effects model framework in the joint model, the index data are used as longitudinal endogenous variables to construct a predictive model for the prognosis of antiretroviral therapy for HIV-infected individuals, so as to dynamically predict the risk of future death.

2. The method for assessing HIV antiretroviral therapy adherence according to claim 1, characterized in that, In S2, the basic form of the group trajectory model GBTM is: In formula (1): Represents a random vector The probability of a random vector Defined as an individual exist Measurement values ​​of variables at each time point The vertical sequence; The group trajectory model GBTM represents the random vectors of individuals. The changing trend of the number of trajectory subgroups in the cluster; For individuals Belongs to the The probability of a group is expressed as the probability of randomly selecting an individual from the population. Belongs to the The probability of the group; Represents the given first Group of random vectors The conditional probability distribution function.

3. The method for assessing HIV antiretroviral therapy adherence according to claim 2, characterized in that, In S2, the individual Belongs to the group probability The calculation is as follows: In formula (2): Indicates the first Group weights.

4. The method for assessing HIV antiretroviral therapy adherence according to claim 3, characterized in that, In S2, the first Group of random vectors conditional probability distribution function The calculation is as follows: In formula (3): Represents an individual Belongs to the The group is at the point in time. Observations The probability distribution function; Represents an individual At the point of time The observed values, namely ART adherence, are reflected by care retention status, medication holding rate, and CD4 cell count; in, This polynomial function is expressed as: In equation (4): Representing continuous time points and observations Latent variables; Represents an individual At the point of time The random error term follows a normal distribution. That is, the mean is 0 and the variance is ; The intercept; The coefficients of the linear terms; The coefficient of the quadratic term; The coefficient of the cubic term; Represents an individual At the point of time The follow-up time is the time represented by the horizontal axis of the trajectory; Based on the range of values ​​for the indicator data, its maximum value is set to... The minimum value is ;like < ,but = ;like > ,but = ;like < < ,but = .

5. The method for assessing HIV antiretroviral therapy adherence according to claim 1, characterized in that, In S4, the Cox proportional hazards model was used to assess the association between each trajectory subgroup and the risk of mortality. Based on the different independent variables included in the Cox proportional hazards model, the following three models were defined to control for confounding effects: The first model is without covariate adjustment, that is, it does not include covariates and only inputs trajectory subgroups; The second model adjusts for demographic variables including age, gender, ethnicity, place of residence, education level, marital status, and body mass index (BMI). The third model is based on the second model and further adjusts the following variables: initial CD4 cell count, ART regimen, whether the ART regimen has been changed, route of infection, drug side effects, whether the patient has received sulfamethoxazole-trimethoprim for the prevention of opportunistic infections, and clinical and treatment variables related to sexually transmitted diseases.

6. The method for assessing HIV antiretroviral therapy adherence according to claim 1, characterized in that, In S5, the random effects model framework in the joint model includes a longitudinal sub-model, which is used to capture the dynamic changes in indicator data to output individual-level trend parameters, wherein: A generalized linear mixed-effects model was used to fit the changes in retention probability over follow-up time for nursing care retention status. A linear mixed-effects model was used to fit the numerical fluctuation trends of drug holding rate and CD4 cell count over time.

7. The method for assessing HIV antiretroviral therapy adherence according to claim 6, characterized in that, In S5, the random effects model framework in the joint model also includes a survival sub-model, which uses the Cox proportional hazards model to correlate the individual-level trend parameters output by the longitudinal sub-model with the risk of death and outputs a risk prediction formula to dynamically predict the risk of future death.

8. A system for assessing HIV antiretroviral therapy adherence, using the method for assessing HIV antiretroviral therapy adherence as described in any one of claims 1 to 7, characterized in that, The system includes: The data acquisition module is used to acquire indicator data reflecting the ART adherence of HIV-infected individuals, including care retention status, medication possession rate, and CD4 cell count. The trajectory grouping module is used to construct the group trajectory model GBTM and, based on the index data, identify trajectory subgroups with different antiretroviral therapy adherence among HIV-infected individuals. The first assessment module is used to calculate the odds ratio (OR) and 95% confidence interval (CI) for each trajectory subgroup using a logistic regression model, in order to assess the impact of different levels of HIV-infected individuals’ demographic and ART variables on the probability of belonging to each trajectory subgroup. The second assessment module is used to calculate the hazard ratio (HR) and 95% confidence interval (CI) for each trajectory subgroup using the Cox proportional hazards model, in order to assess the association between each trajectory subgroup and all-cause mortality, AIDS mortality, and non-AIDS mortality, respectively. The risk prediction module is used to construct a predictive model for the prognosis of antiretroviral therapy for HIV-infected individuals by using the random effects model framework in the joint model as longitudinal endogenous variables, so as to dynamically predict the risk of future death.

9. A device, characterized in that, The device includes: At least one processor; and A memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform a method for assessing HIV antiretroviral therapy adherence as described in any one of claims 1 to 7.

10. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores executable instructions that, when executed, cause the machine to perform the method for assessing HIV antiretroviral therapy adherence as described in any one of claims 1 to 7.

Images