Risk prediction model and construction method for low viremia in HIV-infected patients after antiretroviral therapy

Abstract

The present invention provides a risk prediction model for low viremia in HIV-infected patients after antiretroviral therapy and a method for constructing the model. The model constructed by the present invention has the advantages of being simple, intuitive, and easy to promote and apply, and is helpful for clinicians to provide personalized treatment for HIV-infected patients.

Description

Technical Field

[0001] The present invention relates to the technical field of medical and biological machine learning, and in particular to a risk prediction model for low viremia in HIV-infected persons after antiretroviral therapy and a method for constructing the model. Background Art

[0002] Risk prediction models have been widely used in different diseases to evaluate patient prognosis. The current effective treatment method is to build a mathematical model. At the beginning of modeling, as many known variables as possible are usually selected to overcome the model bias caused by the lack of independent variables. However, in the actual process of building a mathematical model, it is necessary to find some independent variables that have an important impact on the dependent variable. This process is the process of feature selection, eliminating redundant variables. Among them, the filtering method can score each feature, set a threshold or select the number of features to select features, and discard undesirable feature information; the calculation of the Pearson correlation coefficient can also be used to measure the linear correlation between variables, and then obtain the degree of correlation between independent variables and independent variables, and between independent variables and dependent variables; the variance selection method calculates the variance of each feature, and then selects features with variance greater than the threshold according to the threshold. The above methods all use conventional methods to screen known variables and then establish mathematical models, which makes the judgment accuracy of the established mathematical models relatively low. Therefore, it is urgent to introduce new variable selection methods to build effective prediction models to accurately judge the prognosis risk of patients.

[0003] There are many nomogram models for predicting diseases in the prior art, such as:

[0004] CN202010289342.5, the invention name is "Survival probability prediction nomogram for advanced breast cancer, survival probability prediction method and patient classification method", which discloses providing a survival probability prediction nomogram for advanced breast cancer, a survival probability determination method and a patient classification method, wherein the nomogram includes a score scale and also includes several prognostic variables.

[0005] CN202111375000.6, the invention name is "A nomogram prediction system and prediction method for osteoporosis risk in postmenopausal women", which discloses a nomogram prediction system and prediction method for osteoporosis risk in postmenopausal women. Through literature study, a data collection table related to osteoporosis prediction is designed, and a training data set and a validation set are formed through a questionnaire survey and bone density test results of postmenopausal female patients; the training data set is screened by Lasso regression variables, and a prediction model is established by Logistic regression. After verification by establishing a data set and a validation data set, and through model evaluation, the indicators include calibration, resolution and clinical practicality. However, none of the above-mentioned existing patent technologies report a predictive nomogram model for AIDS-related follow-up treatment.

[0006] After people living with HIV (PLWH) receive antiretroviral therapy (ART), most patients can achieve sustained viral suppression. However, some patients still have viral replication that is higher than the detection limit but lower than the criterion for virological failure, which is called low-level viremia (LLV). The World Health Organization defines LLV as a viral load between 50 and 1000 copies / ml after 6 months of ART treatment. The higher the HIV viral load of PLWH patients with LLV, the higher the risk of treatment failure and drug resistance, which will lead to an increased risk of HIV transmission, AIDS morbidity and mortality. Therefore, preventing the occurrence of LLV is crucial for PLWH patients.

[0007] An optimized hepatocellular carcinoma prediction model for chronic hepatitis B with well-controlled viremia. Hye W. Lee disclosed an optimized hepatocellular carcinoma prediction model for chronic hepatitis B with well-controlled viremia.

[0008] Lennemann, Tessa-Suntje (2021): HIV Viral load measurement in the Public Health Approach to HIV / AIDS, published a study of 702 HIV patients receiving ART at 7 research sites, and provided a nomogram prediction model for the risk of virological failure after ART in HIV patients. The outcome predicted by this prediction model is virological failure in HIV patients after ART, with a viral load greater than 400 copies / mL. In addition, this virological failure risk model has obvious shortcomings, including: (1) the small number of people enrolled, which is further divided into a modeling group and a validation group. Each group has a relatively small population, which will affect the identification of key risk factors; (2) R software was not used for rapid variable screening and statistical analysis. The old statistical software has certain limitations in analyzing data, which is time-consuming and of low practicality; (3) variable screening only uses a single method, which is not rigorous enough; (4) the model performance evaluation only uses ROC curve and decision curve evaluation, which is a relatively simple evaluation method.

[0009] Currently, studies have described potential risk factors for low viremia. For example, one study showed that low baseline CD4 + T cell counts and higher baseline VL are associated with the risk of LLV; another study found that other factors associated with LLV include long-term ART, race, and subtype B infection. Although potential risk factors for LLV have been studied, no one factor is decisive, and in most cases, LLV may be caused by the combined action of multiple factors. Therefore, there is a great need to reduce the risk of LLV by mining high-risk factors in clinical test data and conducting risk assessment in advance. The Nomogram prediction model integrates multiple key factors and scores each relevant parameter according to its value. It is a convenient and practical model for clinicians. However, there is currently a lack of research on accurate risk prediction models for LLV in PLWH patients after ART treatment. Summary of the invention

[0010] 1. Technical issues to be resolved

[0011] In view of the above shortcomings and deficiencies of the prior art, the present invention provides a risk prediction model for low viremia in HIV-infected persons after antiretroviral therapy and a method for constructing the model (a risk prediction model for evaluating low viremia in HIV patients after ART treatment and a method for constructing the model). The model helps clinicians to formulate personalized treatment plans for different PLWH patients. At the same time, it solves the technical difficulties of the existing prediction models, such as unreliable models and low recognition rates, caused by too many or too few feature selections.

[0012] (II) Technical solution

[0013] The present invention will provide a risk prediction model for low viremia in HIV-infected patients after antiretroviral therapy and a method for constructing it. According to the WHO definition of low viremia, the viral load is between 50 and 1000 copies / ml after 6 months of ART treatment. The research population included in the present invention includes HIV patients with a viral load of 400 to 999 copies / ml. Compared with the model constructed by Lennemann and Tessa-Suntje (2021), the present invention targets HIV patients with low viral load, and the research is more generalizable. In addition, previous studies have confirmed that HIV patients with low viremia are more likely to develop virological

[0014] Failure, clarifying the risk factors for low viremia, and intervening in the early stage can not only prevent the occurrence of low viremia, but also reduce the incidence of virological failure. At the same time, this model uses R software for rapid statistical analysis, combined with Lasso logistic regression analysis and multivariate logistic regression analysis to jointly identify the key risk factors related to the occurrence of low viremia; ROC curve, calibration curve, decision curve (DCA), clinical impact curve (CIC) are used to comprehensively analyze the predictive efficacy of the low viremia risk model developed in this study from multiple angles.

[0015] The present invention achieves the purpose of the invention through the following technical solutions:

[0016] The first aspect of the present invention is to provide a method for constructing a nomogram model for evaluating the risk of low viremia in HIV-infected patients after receiving antiretroviral therapy, the construction method comprising the following steps:

[0017] S1. Collect patient data, including age, gender, sexual transmission route, marital status, BMI index, antiretroviral ART treatment delay time, baseline CD4 count (baseline CD4 + T cell count), baseline viral load data;

[0018] S2. Data from S1 were preliminarily screened by Lasso logistic regression analysis, and variables with P values ​​less than 0.05 were selected. These variables were then included in the multivariate logistic regression analysis. Variables with P values ​​less than 0.05 were all important influencing factors, and (a) age, (b) ART treatment delay time, and (c) baseline CD4 count were determined to be important clinical data affecting the outcome of low viremia in HIV-infected patients after ART treatment;

[0019] S3. The cutoff value of baseline CD4 count was 300 cells / μL, the cutoff value of ART treatment delay time was 7 months, and the cutoff value of age was 27 years old.

[0020] S4. Based on the important clinical characteristics screened in steps S2 and S3, a nomogram model is constructed to predict the outcome of low viremia in HIV-infected patients after ART treatment.

[0021] The construction method also includes the following steps: S5 evaluating and verifying the model through receiver operating characteristic curve (ROC curve), calibration curve, decision curve (DCA), and clinical impact curve (CIC) analysis.

[0022] In the nomogram, the first line is a score scale with a score range of 0-100, the second line is a baseline CD4 count, 1 corresponds to a baseline CD4 count lower than 300 cells / μl, and 0 corresponds to a baseline CD4 count higher than or equal to 300 / μl; the third line is an ART treatment delay time, 1 corresponds to an ART treatment delay time higher than 7 months, and 0 corresponds to an ART treatment delay time equal to or lower than 7 months; the fourth line is age, 1 corresponds to an age equal to or lower than 27 years old, and 0 corresponds to an age higher than 27 years old; the fifth line is a total risk score with a range of 0-260; the sixth line is the risk of low viremia in PLWH patients after ART treatment, ranging from 0.3-0.5; the risk value formula of the nomogram is: risk value=baseline CD4 count*0.782+ART treatment delay time*0.676+age*0.437.

[0023] The second aspect of the present invention is to provide a nomogram model for predicting the risk of low viremia in HIV-infected patients after receiving antiretroviral therapy established by the above-mentioned construction method. In the nomogram, the first line is a score scale with a score range of 0-100, the second line is a baseline CD4 count, 1 corresponds to a baseline CD4 count of less than 300 cells / microliter, and 0 corresponds to a baseline CD4 count of greater than or equal to 300 cells / microliter; the third line is an ART treatment delay time, 1 corresponds to an ART treatment delay time of greater than 7 months, and 0 corresponds to an ART treatment delay time of equal to or less than 7 months; the fourth line is age, 1 corresponds to an age of equal to or less than 27 years old, and 0 corresponds to an age of greater than 27 years old; the fifth line is a total risk score with a range of 0-260; the sixth line is the risk of low viremia in PLWH patients after receiving ART treatment, with a range of 0.3-0.5; the risk value formula of the nomogram is: risk value=baseline CD4 count*0.782+ART treatment delay time*0.676+age*0.437.

[0024] In detail, the nomogram construction method of the present invention comprises the following steps:

[0025] S1. Collect the patient's baseline clinical data, including age (age when first infected with HIV), gender, sexual transmission route (route of infection with HIV), marital status (marital status when first infected with HIV), body mass index (BMI) (BMI when first infected with HIV, calculated by weight and height according to a formula), and ART treatment delay time (the time interval from when the patient was infected with HIV to when he or she received ART treatment). These five baseline data can be obtained from the patient's medical record system; baseline CD4 + T cell count (baseline CD4 count) can be measured by measuring the CD4 +The number of T cells can be obtained; the baseline viral load (baseline VL) can be obtained by detecting the HIV viral RNA level in the patient's plasma before receiving ART treatment;

[0026] The collected data were first entered into an Excel spreadsheet and then converted into a CSV data format. These data were then imported into SPSS software (version 25.0). SPSS software was then used to organize the data, including (1) data cleaning: deleting duplicate data, deleting rows with missing values ​​for missing data, checking whether the data were within a reasonable range, and removing special characters, spaces, and unnecessary formats from text data; (2) data conversion: converting text data into numerical variables, converting numerical variables into different categories, and creating virtual variables (dummy variables) for categorical variables to facilitate regression analysis; (3) data preparation: determining the variable set and renaming the variables. According to the analysis objectives, relevant variables were selected and irrelevant or redundant variables were excluded.

[0027] S2. Lasso logistic regression analysis and multivariate logistic regression analysis were performed using R statistical software (version 4.1.3). The data of S1 were initially screened by Lasso logistic regression analysis, and the variables with P values ​​less than 0.05 were selected. These variables were then included in the multivariate logistic regression analysis. The variables with P values ​​less than 0.05 were all important influencing factors. It was determined that age, ART delay time, and baseline CD4 count were important clinical data that affected the outcome of LLV in PLWH patients after receiving ART treatment.

[0028] S3, calculated by MedCalc software (version 18.2.1), the baseline CD4 count cutoff value was 300 cells / μL, the ART treatment delay time cutoff value was 7 months, and the age cutoff value was 27 years;

[0029] S4. Based on the important clinical characteristics screened in steps S2 and S3, a nomogram model for predicting the occurrence of LLV outcomes in PLWH patients after ART treatment was constructed using R statistical software (version 4.1.3).

[0030] The present invention also claims a method for evaluating and validating the nomogram: performing receiver operating characteristic curve, calibration curve, decision curve, and clinical impact curve analysis using R statistical software (version 4.1.3) to evaluate and validate the model.

[0031] (III) Beneficial effects

[0032] (1) The present invention uses Lasso logistic regression analysis and multivariate logistic regression analysis to screen characteristic variables from clinical diagnosis, pathology and follow-up information data of HIV-infected patients after antiretroviral therapy, which can effectively find important clinical data for LLV outcomes.

[0033] (2) The present invention provides a risk prediction model for evaluating the occurrence of LLV in PLWH patients after receiving ART treatment. The model constructed by the present invention has the advantages of being simple, intuitive, and easy to promote and apply, which helps clinicians provide personalized treatment for PLWH patients.

[0034] (3) The outcome of this model focuses on the occurrence of low viremia in HIV patients after ART. According to the WHO definition of low viremia, the viral load is between 50 and 1000 copies / ml after 6 months of ART treatment. Therefore, the population targeted by this model includes HIV patients with a viral load of 400 to 999 copies / ml. Compared with previous studies, our study population has more HIV patients with low viral load, and the study is more generalizable. This model included 2145 PLWH for retrospective analysis, of which 1502 were included in the training set and 643 were included in the validation set; R software was used for rapid statistical analysis, and Lasso logistic regression analysis and multivariate logistic regression analysis were combined to jointly identify key risk factors related to the occurrence of low viremia; ROC curves, calibration curves, decision curves (DCA), and clinical impact curves (CIC) were used to comprehensively analyze the predictive efficacy of the low viremia risk model developed in this study. Among the three variables included in this prediction model, age and baseline CD4 count are known variables associated with low viremia, while ART delay time is a newly discovered variable significantly associated with low viremia in this study, suggesting that PLWH patients should start antiretroviral therapy (ART) as early as possible after HIV infection to reduce the possibility of low viremia. Therefore, this model is a new model for predicting low viremia that combines new and old variables, which has good innovation and clinical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The construction and evaluation process of the prediction model in Example 1 includes variable screening, model establishment and model evaluation;

[0036] Figure 2 This is the Lasso logistic regression analysis curve for screening clinical characteristics associated with low viremia in PLWH patients after ART treatment in Example 1;

[0037] Figure 3Comparison of receiver operating characteristic curves of the four prediction models constructed in Example 1;

[0038] Figure 4 is a nomogram of the optimal model constructed in Example 1;

[0039] Figure 5 is the receiver operating characteristic curve of the prediction model training set in Example 1;

[0040] Figure 6 is the receiver operating characteristic curve of the prediction model validation set in Example 1;

[0041] Figure 7 is the calibration curve of the prediction model training set in Example 1;

[0042] Figure 8 is the calibration curve of the prediction model validation set in Example 1;

[0043] Fig. 9 is the decision curve of the prediction model training set in Example 1;

[0044] Fig.10 is the decision curve of the prediction model validation set in Example 1;

[0045] Fig.11 is the clinical impact curve of the training set of the prediction model in Example 1;

[0046] Fig.12 This is the clinical impact curve of the validation set of the prediction model in Example 1. DETAILED DESCRIPTION

[0047] The present invention can be further described by the following examples, however, the scope of the present invention is not limited to the following examples. It will be appreciated by those skilled in the art that various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. The present invention provides general and / or specific descriptions of the materials and test methods used in the test. Although many materials and operating methods used to achieve the purpose of the present invention are well known in the art, the present invention is still described in as much detail as possible at this point.

[0048] The subjects of the present invention were 2152 people living with HIV (PLWH) who received standardized antiretroviral treatment (ART) at Hangzhou Xixi Hospital, the largest designated AIDS hospital in Zhejiang Province, China, between April 5, 2011 and December 29, 2022.

[0049] According to the WHO guidelines, the diagnostic criteria for low viremia (LLV) are: (1) ART treatment for more than 6 months; (2) viral load (VL) between 50 and 1000 copies / mL.

[0050] Inclusion criteria were: (1) age > 18 years; (2) receiving ART for at least 6 months; (3) plasma viral load (VL) less than 1000 copies / mL; and (4) at least two consecutive viremia values ​​per year during the analysis period, with at least 1 month between the two measurements.

[0051] Exclusion criteria were: (1) unknown initial antiretroviral treatment regimen; and (2) incomplete predictor variables.

[0052] In order to establish a multivariate prediction model, according to the inclusion and exclusion criteria, 7 PLWH were excluded due to incomplete clinical data, and a total of 2145 PLWH were included for retrospective analysis. The data set was divided into a training set and a validation set in a ratio of 7:3, with 1502 cases included in the training set and 643 cases included in the validation set ( Figure 1 ), the subsequent analysis process refers to Figure 1 Proceed as shown.

[0053] Statistical methods: SPSS 25.0 and R 4.1.3 were used for statistical analysis. Based on the training set, the least absolute shrinkage and selection operator (LASSO) logistic regression and multivariate logistic regression analysis were used to determine the indicators that can predict LLV, and a nomogram for predicting LLV was established. The receiver operating characteristic curve (ROC) and calibration curve were used to discriminate and calibrate the prediction model. The decision curve analysis (DCA) and clinical impact curve (CIC) were used to evaluate the clinical utility of the prediction model.

[0054] The data of S1 were initially screened by Lasso logistic regression analysis, and the variables with P values ​​less than 0.05 were selected. These variables were then included in the multivariate logistic regression analysis. The variables with P values ​​less than 0.05 were all important influencing factors, and it was determined that (a) age, (b) ART treatment delay time and (c) baseline CD4 count were important clinical data affecting the outcome of low viremia in PLWH patients after ART treatment ( Figure 2 and Table 1).

[0055] Table 1. Multivariate logistic regression analysis

[0056]

[0057] Age, ART delay time and baseline CD4 count (BaselineCD4) were included in four prediction models with different combinations of variables to predict the outcome of low viremia in PLWH patients after ART treatment. Model 1 (model1) was constructed by combining Age and ART delay time; Model 2 (model2) was constructed by combining Age and Baseline CD4; Model 3 (model3) was constructed by combining ART delay time and Baseline CD4; Model 4 (model4) was constructed by combining Age, ART delay time and Baseline CD4. The receiver operating characteristic curve (ROC) analysis was performed and the area under the curve (Areaunder the curve, AUC) was calculated to analyze the predictive ability of the four prediction models (0.618 vs 0.611 vs 0.562 vs 0.632) ( Figure 3 and Table 2), and then through comparative analysis of different models, it was found that Model 4 (Age + ART delay time +Baseline CD4) had significantly better predictive ability than the other three models (Table 3).

[0058] Table 2. Analysis of prediction ability of different models

[0059]

[0060] Table 3. Analysis of the difference in prediction ability of different models

[0061]

[0062] A nomogram model was constructed to predict the outcome of low viremia in PLWH patients after ART treatment based on age, ART treatment delay time and baseline CD4 count ( Figure 4 ).

[0063] In the nomogram, the first line is a score scale with a score range of 0-100, the second line is the baseline CD4 count, 1 corresponds to a baseline CD4 count lower than 300 cells / μl, and 0 corresponds to a baseline CD4 count higher than or equal to 300 / μl; the third line is the ART treatment delay time, 1 corresponds to an ART treatment delay time higher than 7 months, and 0 corresponds to an ART treatment delay time equal to or lower than 7 months; the fourth line is the age, 1 corresponds to an age equal to or lower than 27 years old, and 0 corresponds to an age higher than 27 years old; the fifth line is the total risk score, ranging from 0-260; the sixth line is the risk of low viremia in PLWH patients after ART treatment, ranging from 0.3-0.5.

[0064] Application example: Figure 4 As shown by the red arrow, one of the PLWH patients had a baseline CD4 count greater than 300 cells / μL, a delay in ART treatment of more than 7 months, and was younger than 27 years old. The corresponding risk values ​​were 0, 87, and 57, respectively. The total risk score was 144, and the corresponding risk of low viremia was 0.34, which means that the PLWH patient has a 34% probability of developing low viremia after receiving ART treatment.

[0065] In order to illustrate the predictive ability of the nomogram, the training set ( Figure 5 and Table 4) and validation set ( Figure 6 Receiver operating characteristic curve (ROC) analysis was performed and the area under the curve (AUC) was calculated, which showed good predictive ability (0.632 vs 0.594).

[0066] Table 4. Analysis of the area under the curve (AUC) value of the training set

[0067]

[0068] Table 5. Analysis of the area under the curve (AUC) value of the validation set

[0069]

[0070] Figure 7 and Figure 8 is the calibration curve with 1000 Bootstrap resamplings. The results show that both the training set ( Figure 7 ) or validation set ( Figure 8 ) The predictive ability of the model is in good agreement with the actual observation results. Fig. 9 and Fig.10 The results show that the net benefit of the nomogram is in the training set ( Fig. 9 ) and validation set ( Fig.10 ) has a threshold range of 0.2-0.6. Fig.11 is the clinical impact curve, and the results show that the training set ( Fig.11) and validation set ( Fig.12 ) When the risk threshold is greater than 0.5, the model prediction and actual occurrence are highly matched, and the clinical prediction efficiency is high.

[0071] The above results all indicate that the nomogram constructed by three clinically easily available characteristic data, namely age, ART delay time and baseline CD4 count, has good clinical predictive ability and clinical applicability.

[0072] Among the three variables included in this prediction model, age and baseline CD4 count are known variables associated with low viremia, while ART delay time is a newly discovered variable significantly associated with low viremia in this study, suggesting that PLWH patients should start antiretroviral therapy (ART) as early as possible after HIV infection to reduce the possibility of low viremia. Therefore, this model is a new model for predicting low viremia that combines new and old variables, which has good innovation and clinical significance.

[0073] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-restrictive from any point of view, and the scope of the present invention is defined by the technical solution rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the technical solution are included in the present invention.

Claims

1. A method for constructing a nomogram model for predicting the risk of hypoviremia in HIV-infected patients after receiving antiretroviral therapy, characterized in that: The construction method comprises the following steps: S1. Collect patient data, including age, gender, sexual transmission route, marital status, BMI index, antiretroviral ART treatment delay time, baseline CD4 count, and baseline viral load data; S2. The data of S1 were initially screened by Lasso logistic regression analysis, and the variables with P values ​​less than 0.05 were selected. These variables were then included in the multivariate logistic regression analysis. The variables with P values ​​less than 0.05 were all important influencing factors, and it was determined that age, ART treatment delay time and baseline CD4 count were important clinical data affecting the outcome of low viremia in HIV-infected patients after ART treatment; S3. The cutoff value of baseline CD4 count was 300 cells / μL, the cutoff value of ART treatment delay time was 7 months, and the cutoff value of age was 27 years old. S4, constructing a nomogram model for predicting the outcome of low viremia in HIV-infected patients after ART treatment based on the important clinical characteristics screened in steps S2 and S3; In the nomogram, the first line is a score scale with a score range of 0-100, the second line is a baseline CD4 count, 1 corresponds to a baseline CD4 count lower than 300 cells / μl, and 0 corresponds to a baseline CD4 count higher than or equal to 300 cells / μl; the third line is an ART treatment delay time, 1 corresponds to an ART treatment delay time higher than 7 months, and 0 corresponds to an ART treatment delay time equal to or lower than 7 months; The fourth line is age, 1 corresponds to age equal to or less than 27 years old, and 0 corresponds to age greater than 27 years old; The fifth line is the total risk score, ranging from 0 to 260; the sixth line is the risk of low viremia in PLWH patients after ART treatment, ranging from 0.3 to 0.5; the risk value formula of the nomogram is: risk value = baseline CD4 count*0.782+ART treatment delay time*0.676+age*0.

437.

2. The method for constructing a risk prediction nomogram model according to claim 1, characterized in that: The construction method also includes the following steps: evaluating and verifying the model through receiver operating characteristic curve, calibration curve, decision curve, and clinical impact curve analysis.

Images