Continuous medical index and clinical outcome optimal association boundary value determination model and application

Abstract

The invention relates to the technical field of medical statistics and clinical information processing, in particular to a data-driven model for determining an optimal association boundary value of a continuous medical index and a dichotomy clinical outcome and application of the data-driven model. The method is driven by data, the bias of subjective experience judgment is avoided, and the most significant statistical critical value associated with the clinical outcome is automatically and objectively found. A large number of candidate critical values are traversed, non-linear fitting is carried out in combination with a restrictive cubic spline, the lowest point of the p value on a fitting curve is accurately positioned, the accumulation histogram is assisted to visually present the result distribution difference corresponding to different threshold values, the optimal correlation boundary value is finally determined, and the precision of the optimal correlation boundary value is remarkably higher than that of a traditional grouping method. The method can be applied to correlation analysis of other continuous medical indexes (such as CD4 counting, virus load, biochemical indexes and the like) and various binary classification clinical outcomes (such as survival states, treatment reactions, complications and the like), and does not depend on specific diseases or indexes. According to the method, visual chart output can be integrated, so that the result is more visual, and clinicians can understand and apply the result conveniently.

Description

Technical Field

[0001] This patent relates to the field of medical statistics and clinical information processing technology, specifically to a data-driven model for determining the optimal correlation threshold between continuous medical indicators and binary clinical outcomes and its application. Background Technology

[0002] Tuberculous meningitis (TBM) is a type of tuberculosis of the central nervous system caused by Mycobacterium tuberculosis, which primarily infects the lungs, and then invades the subarachnoid space via the bloodstream or directly, resulting in occult infection of the meninges (Sanjay K Jain, David M Tobin, Elizabeth W Tucker, et al. Tuberculouseningitis: a roadmap for advancing basic and translational research. NatImmunol. 2018;19(6):521-525.). It is the most severe manifestation of Mycobacterium tuberculosis infection and has the highest mortality rate among tuberculosis types. In 2019, of the 7.1 million newly diagnosed tuberculosis cases worldwide, approximately 164,000 were adults with tuberculous meningitis (TBM), with about a quarter of these cases occurring in individuals living with HIV (Dodd PJ, Osman M, Cresswell FV, et al. The global burden of tuberculous meningitis in adults: A modellingstudy. PLOS Glob Public Health. 2021;1(12):e0000069. Published 2021 Dec 8.). Co-infected individuals often face worse outcomes (Boonyagars L, Sangketchon C, Pholtawornkulchai K. Presentation, Clinical Characteristics, and Treatment Outcomes among Tuberculous Meningitis Patients with and Without HIV Infection at Vajira Hospital, Thailand: A Retrospective Cohort Study). J Int Assoc Provid AIDS Care. 2021;20:23259582211045551. ;Thwaites GE, Bang ND, Dung NH, et al. Dexamethasone for the treatment of tuberculous meningitis in adolescents and adults[J]. New England Journal of Medicine, 2004, 351(17): 1741-1751.). Although the overall mortality rate of TBM is about 25%, the mortality rate of HIV co-infected individuals increases to nearly 50% (Stadelman AM, Ellis J, Samuels THA, et al. Treatment Outcomes in Adult Tuberculous Meningitis: A Systematic Review and Meta-analysis. Open Forum Infect Dis. 2020;7(8):ofaa257.). Severe immunosuppression, especially CD4 + T lymphocyte depletion is a hallmark of HIV infection and a key driver of susceptibility to tuberculosis. HIV-infected individuals with CD4 counts below 200 cells / μL have a 25 times higher risk of developing tuberculosis than those with CD4 counts above 1000 cells / μL (Ellis PK, Martin WJ, Dodd PJ. CD4 count and tuberculosis risk in HIV-positive adults not on ART: a systematic review and meta-analysis. Peer J. 2017;5:e4165. Published 2017 Dec 14.). A study in Vietnam involving 70 HIV-infected individuals with tuberculous meningitis showed that their median CD4 count was only 67 cells / μL (IQR 19-124), highlighting the degree of immunodeficiency in this population (Hai HT, Thanh Hoang Nhat L, Tram TTB, et al. Whole blood transcriptional profiles and the pathogenesis of tuberculousmeningitis. Elife. 2024;13:RP92344. Published 2024 Oct 30.).

[0003] CD4, a continuous medical indicator + T lymphocyte count is a key influencing factor in binary clinical outcomes (e.g., death / survival, improvement / deterioration). Determining an optimal cutoff value is crucial for disease stratification, prognosis, treatment decisions, and guideline development. Currently, methods for determining such cutoff values ​​largely rely on clinical experience, reference to previous literature, or statistical distributions. However, these methods suffer from high subjectivity, failure to fully utilize the correlation between the data itself and the outcome, and inability to accurately identify the point of strongest statistical association. The World Health Organization defines a CD4 count <50 cells / μL as "advanced HIV disease," a stage characterized by immune system collapse and near-total loss of defense capabilities, leading to a series of extremely serious consequences (Guidelines for Managing Advanced HIV Disease and Rapid Initiation of Antiretroviral Therapy. Geneva: World Health Organization; 2017.). Studies have shown that low CD4 counts are associated with a higher risk of adverse outcomes (Loghin II, Vâță A, Miftode EG, et al. Characteristics of Tuberculous Meningitis in HIV-Positive Patients from Northeast Romania. Clin Pract. 2023;13(6):1488-1500. Published 2023 Nov 21.). However, whether the aforementioned threshold is the strongest association with HIV combined with TBM still requires verification with a larger dataset. Current technology lacks an automated, objective, universal, and accurate method for calculating the optimal association threshold between continuous medical indicators and binary clinical outcomes, leading to inaccurate disease risk stratification and subjective biases in treatment decisions. Therefore, there is an urgent need for a patent application for a model that can automatically and accurately identify the statistically most statistically associated threshold with clinical outcomes, driven by the data itself. Summary of the Invention

[0004] This invention discovers that a data-driven model can automatically and accurately determine the CD4 prognosis of HIV combined with TBM. + The invention utilizes a T lymphocyte count threshold to test a series of consecutive critical values, thereby improving the accuracy of the best correlation threshold after prediction. Based on this, the invention was completed.

[0005] In a first aspect, the present invention provides a CD4-based method for automatically and accurately determining the prognosis of HIV combined with TBM. +A method for setting T lymphocyte count cutoff values, the method comprising the following steps: S1. Initial Dataset: Input the patient's baseline CD4 count and binary outcome data to form the initial dataset; S2. Data Preprocessing: Examine the distribution of continuous independent variables, observe the data distribution status, determine the method for removing extreme values, and remove extreme values; S3. Generate candidate critical values: Systematically generate a series of candidate critical values ​​from the data processed in S2 in steps of 0.5 units. S4. Traverse and calculate associativity p Values: Using each candidate cutoff value as a boundary, patients are divided into two groups: those with CD4 greater than or equal to the threshold and those with CD4 less than the threshold. A 2x2 contingency table is constructed to correlate these groups with the binary outcome. Based on the expected frequency of each cell in the contingency table, a series of consecutive cutoff values ​​are tested one by one to calculate the significance of the association between the group and the binary outcome at each cutoff value. p value; S5. Based on Restricted Cubic Spline (RCS) p The trend of the value changing with the threshold is non-linearly fitted: each p The values ​​are represented as scatter points in the graph, and their RCS curves are plotted. S6. Draw a stacked bar chart to show the differences in the distribution of outcomes corresponding to different thresholds, and then determine the optimal association threshold; the output of the optimal association threshold is the result.

[0006] Furthermore, in step S4, the test is selected from the chi-square test or Fisher's exact test.

[0007] Furthermore, when any expected frequency in the contingency table is less than 5, Fisher's exact test is used; otherwise, the chi-square test is used. Each of the generated series of continuous critical values ​​is tested to improve the accuracy of predicting the optimal association threshold.

[0008] Furthermore, in step S6, the optimal correlation threshold is defined as the value on the RCS fitting curve. p The point with the smallest global value.

[0009] Secondly, the present invention provides a CD4-based method for automatically and accurately determining the prognosis of HIV combined with TBM. + A system for T lymphocyte count cutoffs, the system comprising the following modules: Data input module: Input the patient's baseline CD4 count and binary outcome data to form the initial dataset; Data preprocessing module: Examine the distribution of continuous independent variables, observe the data distribution, determine methods for removing extreme values; remove initial data. Concentrated extreme values; Data processing module: The preprocessed data is used to systematically generate a series of candidate critical values ​​in steps of 0.5 units. Using each candidate cutoff value as a boundary, patients were divided into two groups: those with CD4 levels greater than or equal to the threshold and those with CD4 levels less than the threshold. A 2x2 contingency table was constructed to correlate these groups with the one-year mortality outcome. Based on the expected frequencies of the cells in the contingency table, a series of consecutive cutoff values ​​were tested to calculate the significance of the association between the group and the binary outcome at each cutoff value. p value; Based on restricted cubic spline (RCS) p The trend of the value changing with the threshold is non-linearly fitted: each p The values ​​are represented as scatter points in the graph, and their RCS curves are plotted. The optimal association threshold is determined by displaying the differences in outcome distribution corresponding to different thresholds using stacked bar charts. Data output module: Output the optimal correlation threshold.

[0010] Furthermore, in the data processing module, the test is selected from the chi-square test or Fisher's exact test.

[0011] Furthermore, when any expected frequency in the contingency table is less than 5, Fisher's exact test is used; otherwise, the chi-square test is used. The generated series of continuous critical values ​​are tested to improve the accuracy of predicting the optimal association critical value.

[0012] Furthermore, in the data processing module, the optimal correlation threshold is defined as the value on the RCS fitting curve. p The point with the smallest global value.

[0013] Beneficial effects Compared with the prior art, the beneficial effects of the present invention are as follows: 1) Objectivity and data-driven: This invention is data-driven, avoiding the bias of subjective experience judgment, and can automatically and objectively find the statistical critical value that is most significantly associated with clinical outcomes.

[0014] 2) High accuracy: By traversing a large number of candidate critical values ​​and combining them with restrictive cubic splines for nonlinear fitting, it can accurately locate the fitted curve. p By finding the lowest value, the optimal correlation boundary can be found, with a much higher accuracy than traditional grouping methods.

[0015] 3) High versatility: This method can be applied to the association analysis of other continuous medical indicators (such as CD4 count, viral load, biochemical indicators, etc.) with various binary clinical outcomes (such as survival status, treatment response, occurrence of complications, etc.), without relying on specific diseases or indicators. 4) Visual output: The method can integrate visual charts and graphs to make the results more intuitive and easier for clinicians to understand and apply. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 To determine the prognosis of HIV combined with TBM, CD4 + Flowchart of the method for finding the optimal correlation threshold for T lymphocyte count.

[0018] Figure 2 Record the initial data for input.

[0019] Figure 3 To view the distribution results of continuous independent variables.

[0020] Figure 4 After data preprocessing, a series of candidate critical value results are systematically generated in steps of 0.5 units.

[0021] Figure 5 To calculate the associativity during traversal p Value operation process.

[0022] Figure 6 To make each p The values ​​are represented as scatter points in the graph, and the RCS curve is plotted.

[0023] Figure 7 The optimal CD4 threshold was determined for the discharge outcome of 201 HIV patients with TBM.

[0024] Note: Blue dots: Fisher's exact test; purple dots: Chi-square test; hollow dots represent... p <0.05; Red line is the RCS curve; X-axis: CD4 + T lymphocyte count; Left Y-axis: pValue; Right Y-axis: Number of patients; Prediction area: Yellow, correctly predicted deterioration; Pink, incorrectly predicted deterioration; Light blue, correctly predicted improvement; Green, incorrectly predicted improvement.

[0025] Figure 8 The optimal CD4 threshold was used to determine the discharge outcomes of 447 HIV patients with TBM.

[0026] Note: Blue dots: Fisher's exact test; purple dots: Chi-square test; hollow dots represent... p <0.05; Red line is the RCS curve; X-axis: CD4 + T lymphocyte count; Left Y-axis: p Value; Right Y-axis: Number of patients; Prediction area: Yellow, correctly predicted deterioration; Pink, incorrectly predicted deterioration; Light blue, correctly predicted improvement; Green, incorrectly predicted improvement.

[0027] Figure 9 ROC curves for 447 HIV patients with TBM. Detailed Implementation

[0028] The embodiments of the present invention will be described in detail below with reference to examples. However, those skilled in the art will understand that the following examples are for illustrative purposes only and should not be considered as limiting the scope of the invention. Unless otherwise specified in the examples, conventional conditions or conditions recommended by the manufacturer shall apply.

[0029] Example 1: Complete Process 1. Input the initial dataset Input the patient's baseline CD4 count and binary outcome data to form the initial dataset; obtain the required basic variables (such as CD4 baseline value, discharge outcome / one-year mortality outcome, etc.).

[0030] 2. Examine the distribution of continuous independent variables. Observe the data distribution to determine the method for removing extreme values; observe the overall distribution of CD4 counts (whether it is normal, skewed, or has extreme values); extreme values ​​(abnormally high or low) affect the stability of subsequent optimal correlation threshold selection; judging the data distribution pattern helps to decide whether to perform transformation or tail reduction processing.

[0031] Extreme values ​​arise from measurement errors, rare clinical situations (such as extremely high CD4 levels but poor prognosis), and are not representative of the population; eliminating extreme values ​​helps to eliminate interference in the process of determining the optimal association threshold.

[0032] 3. Data Preprocessing Remove extreme values ​​from the initial data; 4. Generate candidate critical values The processed data is systematically processed in steps of 0.5 units to generate a series of candidate cutoff values. All possible CD4 cutoff values ​​are systematically examined. If only a few values ​​are selected based on clinical experience (such as 200 or 350), the best prediction point may be missed. By generating multiple candidate values ​​within a reasonable range with a smaller step size (0.5), the boundary point with the greatest difference in risk of outcome can be captured more precisely.

[0033] 5. Traverse and calculate associations p value Using each candidate cutoff value as a boundary, patients were divided into two groups: those with CD4 levels greater than or equal to the threshold and those with CD4 levels less than the threshold. A 2x2 contingency table was constructed to correlate these groups with one-year mortality outcomes. Based on the expected frequencies of the cells in the contingency table, a series of consecutive cutoff values ​​were tested to calculate the significance of the association between the group and mortality outcomes at each cutoff value. p Value; Calculation p The value reflects the statistical strength of the association between the group and the outcome.

[0034] Fisher's exact test is used when any expected frequency in the contingency table is less than 5; otherwise, the chi-square test is used.

[0035] 6. Fitting based on restricted cubic spline (RCS) p Value curve original p Value sequences may fluctuate; RCS curves can capture their non-linear trends and reduce noise from random errors. They facilitate visualization and precise reading of minimum values. p The threshold corresponding to the value. p The trend of the value changing with the threshold is non-linearly fitted; each p The values ​​are represented as scatter points in the graph; plot their RCS curve; smooth. p The scatter plot of values ​​changing with the threshold can be found more stably. p The lowest point of the value.

[0036] 7. Draw a stacked bar chart to determine the optimal correlation threshold. By combining RCS curves and using stacked bar charts to illustrate the differences in outcome distribution corresponding to different thresholds, the optimal association cutoff value is determined. Based on statistical significance and clinical interpretability, the final threshold is determined. The optimal association cutoff value is the result.

[0037] like Figure 1 As shown, CD4 counts are used to determine the prognosis of HIV combined with TBM. + Flowchart of the method for finding the optimal correlation threshold for T lymphocyte count. Figure 2 Record the initial data for input. Figure 3 To view the distribution results of continuous independent variables. Figure 4 After data preprocessing, a series of candidate critical value results are systematically generated in steps of 0.5 units. Figure 5 To calculate the associativity during traversal p Value operation process. Figure 6 To make each p The values ​​are represented as scatter points in the graph, and the RCS curve is plotted.

[0038] Example 2: CD4 counts of discharge outcomes in 201 HIV-positive patients with TBM + Optimal correlation threshold for T lymphocyte count (1) Study population and data sources This study tested data from a multicenter study of HIV patients with tuberculous meningitis. The study recruited participants from eight tertiary infectious disease hospitals in eight provinces of China between January 2018 and December 2020: Shenzhen (n=7 patients); Shanghai (n=9 patients); Chongqing (n=103 patients); Tianjin (n=4 patients); Fuzhou (n=16 patients); Zhengzhou (n=31 patients); Hangzhou (n=2 patients); and Nanning (n=29 patients).

[0039] Inclusion criteria: HIV diagnosis was based on the "Guidelines for the Diagnosis and Treatment of HIV in China (2024 Edition)," and tuberculous meningitis diagnosis was based on the "Clinical Guidelines for Central Nervous System Tuberculosis in China (2023 Edition)." Inpatients aged 18 years and older with confirmed HIV testing and a clinical diagnosis of TBM were included.

[0040] Ultimately, a total of 201 HIV patients with TBM who met the above criteria were included in this analysis.

[0041] (2) Variable definition and data collection Exposure variable: Peripheral blood CD4 counts were collected from all patients with HIV and TBM at admission. + T lymphocyte count (unit: cells / μL).

[0042] Outcome variable: Defined as the degree of improvement / deterioration of the patient at discharge.

[0043] (3) Steps for applying the method of the present invention S1: Data Acquisition and Preparation: Baseline CD4 counts and discharge status data of the above 201 patients were extracted from the hospital's electronic medical record system to form the initial dataset.

[0044] S2: Data preprocessing: To reduce the impact of extreme values ​​on model stability, CD4 counts are preprocessed. The 10th percentile (P10) and 90th percentile (P90) of the CD4 counts (skewed) in this data are calculated, and extreme values ​​below P10 and above P90 are excluded. Only the middle 80% of the data range is analyzed.

[0045] S3: Generate candidate critical values. Within the preprocessed CD4 count range, a series of candidate critical values ​​are systematically generated in steps of 0.5 units.

[0046] S4: Traverse and calculate associativity p For each candidate cutoff value (e.g., 50 CD4 / μL): patients were divided into two groups: a CD4 count < 50 / μL group and a CD4 count ≥ 50 / μL group. A 2x2 contingency table of this grouping and one-year mortality outcomes was constructed.

[0047] Based on the expected frequencies of the cells in the contingency table, the chi-square test or Fisher's exact test is automatically selected to calculate the significance of the association between the group and the outcome at the critical value. p value.

[0048] S5: Fitting and Determining the Optimal Correlation Boundary Value Based on restricted cubic spline (RCS) p The trend of the value changing with the threshold is nonlinearly fitted, and the difference in the distribution of the outcome corresponding to different thresholds is shown based on the stacked bar chart, thereby determining the optimal association threshold.

[0049] (4) Results like Figure 7 As shown, CD4 + A T lymphocyte count of <50 cells / μL is the optimal threshold for discharge outcomes in HIV patients with TBM.

[0050] Example 3: CD4 counts of 447 HIV patients with TBM upon discharge. + Optimal correlation threshold for T lymphocyte count (1) Study population and data sources This study collected data from 447 adult hospitalized patients diagnosed with tuberculosis melanoma (TBM) and co-infected with HIV between January 2018 and December 2024, all of whom received anti-tuberculosis treatment. The patients were from seven tertiary infectious disease hospitals in China: Guangzhou (22 cases), Nanning (21 cases), Shanghai (38 cases), Zhengzhou (15 cases), Chongqing (53 cases), Shenzhen (8 cases), and Chengdu (290 cases).

[0051] Inclusion criteria: HIV diagnosis was based on the "Chinese Guidelines for the Diagnosis and Treatment of HIV / AIDS (2024 Edition)," and tuberculous meningitis diagnosis was based on the "Chinese Clinical Guidelines for Central Nervous System Tuberculosis (2023 Edition)." Adult hospitalized patients with confirmed HIV testing and a clinical diagnosis of tuberculous meningitis were included.

[0052] Ultimately, a total of 447 HIV patients with TBM who met the above criteria were included in this analysis.

[0053] (2) Variable definition and data collection Exposure variable: Peripheral blood CD4 counts were collected from all patients at admission. + T lymphocyte count (unit: cells / μL).

[0054] Outcome variables: When defining binary clinical outcomes, the modified Rankin Scale (mRS score) was used to assess discharge status. mRS 0-2 was considered good prognosis / functional independence, and mRS 3-6 was considered poor prognosis / functional dependence or death.

[0055] (3) Steps for applying the method of the present invention S1: Data Acquisition and Preparation: Baseline CD4 counts and discharge status data of the above 447 patients were extracted from the hospital's electronic medical record system to form the initial dataset.

[0056] S2: Data preprocessing: To reduce the impact of extreme values ​​on model stability, CD4 counts are preprocessed. The 10th percentile (P10) and 90th percentile (P90) of the CD4 counts (skewed) in this queue are calculated, and extreme values ​​below P10 and above P90 are excluded. Only the middle 80% of the data range is analyzed.

[0057] S3: Generate candidate critical values. Within the preprocessed CD4 count range, a series of candidate critical values ​​are systematically generated in steps of 0.5 units.

[0058] S4: Traverse and calculate associativity p For each candidate cutoff value (e.g., 50 CD4 counts / μL): patients are divided into two groups: a CD4 count < 50 CD4 counts / μL group and a CD4 count ≥ 50 CD4 counts / μL group. A 2x2 contingency table is constructed showing the association between this grouping and discharge outcome. Based on the expected frequencies of the cells in the contingency table, either a chi-square test or Fisher's exact test is automatically selected to calculate the significance of the association between this grouping and discharge outcome at the cutoff value. p value.

[0059] S5: Fitting and Determining the Optimal Correlation Boundary Value Based on restricted cubic spline (RCS)p The trend of the value changing with the threshold is nonlinearly fitted, and the difference in the distribution of the outcome corresponding to different thresholds is shown based on the stacked bar chart, thereby determining the optimal association threshold.

[0060] (4) Results The results showed that CD4 + A T lymphocyte count of <50 cells / μL is the optimal cutoff value for discharge outcomes in HIV-positive patients with TBM (see [reference needed]). Figure 8 The ROC curve shows CD4 + T lymphocyte count has a good predictive ability for patient discharge status. p =0.001, AUC=0.618, 95% confidence interval is 0.549-0.686). At this point, using CD4 count <50 cells / μL as the cutoff value, the sensitivity is 68.5%, specificity is 74.1%, and accuracy is 73.2%, which is a good result.

Claims

1. An automated and accurate CD4 prognosis determination method for HIV combined with TBM + A method for setting T lymphocyte count cutoff values, the method comprising the following steps: S1. Initial Dataset: Input the patient's baseline CD4 count and binary outcome data to form the initial dataset; S2. Data Preprocessing: Examine the distribution of continuous independent variables: Observe the data distribution, determine the method for removing extreme values, and remove extreme values; S3. Generate candidate critical values: Systematically generate a series of candidate critical values ​​from the data processed in S2 in steps of 0.5 units. S4. Traverse and calculate associativity p Values: Using each candidate cutoff value as a boundary, patients are divided into two groups: those with CD4 levels greater than or equal to the threshold and those with CD4 levels less than the threshold. A 2x2 contingency table is constructed to correlate these groupings with the outcome. Based on the expected frequencies of the cells in the contingency table, a series of cutoff values ​​are tested to calculate the significance of the association between the grouping and the outcome at each cutoff value. p value; S5. Based on Restricted Cubic Spline (RCS) p The trend of the value changing with the threshold is non-linearly fitted: each p The values ​​are represented as scatter points in the graph, and their RCS curves are plotted. S6. Draw a stacked bar chart to show the differences in the distribution of outcomes corresponding to different thresholds, and then determine the optimal association threshold; the output of the optimal association value is the result.

2. The method as described in claim 1, wherein in step S4, the test is selected from the chi-square test or Fisher's exact test; when any expected frequency in the contingency table is less than 5, Fisher's exact test is used; otherwise, the chi-square test is used; and the generated series of continuous critical values ​​are tested one by one to improve the accuracy of predicting the best correlation critical value.

3. The method as described in claim 1, wherein in step S6, the optimal critical value is on the RCS fitting curve. p The point with the smallest global value.

4. The method as described in claim 1, characterized in that: All analyses were performed in R software. Plotting was done using the ggplot2 package, data preparation was done using the dplyr and tidyr packages, and restricted cubic spline fitting was done using the splines package.

5. An automated and accurate CD4 prognosis determination method for HIV combined with TBM + A system for T lymphocyte count cutoffs, the system comprising the following modules: Data input module: Input the patient's baseline CD4 count and binary outcome data to form the initial dataset; Data preprocessing module: View the distribution of continuous independent variables: Observe the data distribution and determine the method to remove extreme values; Remove initial data Extreme values; Data processing module: The preprocessed data is used to systematically generate a series of candidate critical values ​​in steps of 0.5 units. Using each candidate cutoff value as a boundary, patients were divided into two groups: those with CD4 levels greater than or equal to the threshold and those with CD4 levels less than the threshold. A 2x2 contingency table was constructed to correlate these groups with the one-year mortality outcome. Based on the expected frequency of each cell in the contingency table, a series of consecutive cutoff values ​​were tested one by one to calculate the significance of the association between the group and the binary outcome at each cutoff value. p value; Based on restricted cubic spline (RCS) p The trend of the value changing with the threshold is non-linearly fitted: each p The values ​​are represented as scatter points in the graph, and their RCS curves are plotted. The optimal association threshold is determined by displaying the differences in outcome distribution corresponding to different thresholds using stacked bar charts. Data output module: Output the optimal correlation threshold.

6. In the system as described in claim 5, the test in the data processing module is selected from the chi-square test or Fisher's exact test; when any expected frequency in the contingency table is less than 5, Fisher's exact test is used; otherwise, the chi-square test is used; the generated series of continuous critical values ​​are tested to improve the accuracy of predicting the optimal correlation boundary value.

7. In the system of claim 5, in the data processing module, the optimal correlation threshold is the value on the RCS fitting curve. p The point with the smallest global value.

Images