A method for constructing a prognosis evaluation model based on HIV infection sepsis

Abstract

The application discloses a kind of based on HIV infection sepsis prognosis evaluation model construction method, comprising the following steps: obtaining the sample data and corresponding clinical information of HIV infection sepsis patient, construct derivation cohort, internal verification cohort and external verification cohort;Based on derivation cohort, screening is carried out by LASSO-Cox regression and 10-fold cross validation is used, to obtain candidate predictor;Candidate predictor is eliminated by recursive feature, and key prediction variable is identified;After classification to key prediction variable, input multivariate Cox proportional hazards model, and output independent prediction variable and its corresponding regression coefficient;According to regression coefficient, the risk score of each independent prediction variable is obtained and is accumulated, to obtain prognosis risk score;The discrimination, calibration degree and clinical utility of prognosis evaluation model are evaluated.The application comprehensively covers the multiple variables of HIV related sepsis, and improves the prediction accuracy of HIV related sepsis.

Description

Technical Field

[0001] This invention relates to the field of sepsis prognostic scoring, and in particular to a method for constructing a prognostic assessment model for HIV-infected sepsis. Background Technology

[0002] Sepsis is a leading cause of death and healthcare depletion worldwide. In individuals infected with Human Immunodeficiency Virus (HIV), the incidence and mortality rates of sepsis are significantly higher than in the general population. The interaction between HIV-associated immunodeficiency, chronic immune activation, and opportunistic infections produces specific sepsis phenotypes characterized by significant heterogeneity, rapid organ dysfunction, and highly variable outcomes. Therefore, early and accurate risk stratification at patient admission is crucial for optimizing resource allocation and adjusting the timing and intensity of interventions. Existing prognostic assessment models, such as the Sequential Organ Failure Assessment (SOFA), Acute Physiology and Chronic Health Assessment II (APACHE II), and Rapid Sequential Organ Failure Assessment (qgSOFA), are primarily developed based on HIV-negative individuals and may fail to adequately reflect the unique immunological and virological characteristics of HIV and its complex, non-linear interactions with multi-organ damage. In practice, the general scoring criteria of these prognostic assessment models often misclassify risk categories in HIV-infected individuals, thus limiting their applicability in clinical decision-making and clinical trial enrollment. The prognosis of HIV-associated sepsis is influenced by a variety of factors, including the degree of immune dysfunction, viral load, opportunistic infections, and adherence to and resistance to antiretroviral drugs. These factors are not fully covered by traditional prognostic assessment models, resulting in poor performance of existing prognostic assessment models in HIV-infected sepsis patients. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings and deficiencies of the prior art and provide a method for constructing a prognostic assessment model for HIV-infected sepsis.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A method for constructing a prognostic assessment model for HIV-infected sepsis includes the following steps:

[0006] S1. Constructing the dataset: Obtain sample data and corresponding clinical information of HIV-infected sepsis patients, and construct the derivation queue, internal validation queue and external validation queue;

[0007] S2. Prognostic Assessment Model Construction: Based on the derivation queue, candidate predictors are obtained through screening using minimum absolute contraction and the Cox regression selection operator, followed by 10-fold cross-validation. Candidate predictors are then eliminated using recursive features to identify key predictors. After classifying key predictors according to clinical relevance thresholds, they are input into a multivariate Cox proportional hazards model, which outputs independent predictors and their corresponding regression coefficients. Each regression coefficient is multiplied by 10 and rounded to the nearest integer to obtain a risk score for each independent predictor. The risk scores of the independent predictors are then summed to obtain the prognostic risk score.

[0008] The independent predictors included age, ongoing renal replacement therapy, septic shock, HIV viral load, international normalized ratio, lactate dehydrogenase, lactate, blood urea nitrogen, and albumin.

[0009] S3. Model Evaluation: The prognostic assessment model was evaluated for discrimination, calibration and clinical utility using the derivation cohort, internal validation cohort and external validation cohort respectively.

[0010] Step S1 involves obtaining sample data and corresponding clinical information from HIV-infected sepsis patients, and screening the sample data. The screening criteria for the sample data are: patients are adults who are diagnosed with HIV infection during hospitalization and meet the sepsis-3 criteria; patients who do not meet the criteria are excluded, and the exclusion criteria include: missing key laboratory data; age <18 years; hospitalization time <24 hours; pregnancy; laboratory tests not completed within 48 hours of the onset of sepsis.

[0011] Based on the sample data obtained after screening, an inference queue, an internal validation queue, and an external validation queue are constructed. The inference queue and the internal validation queue are from the hospital's electronic health record system. The inference queue consists of patients admitted between January 2013 and December 2023, and the internal validation queue consists of patients admitted between January 2024 and March 2025.

[0012] The external validation cohort consists of two independent cohorts: the MIMIC-IV cohort and the eICU-CRD cohort. The MIMIC-IV cohort consists of patients selected from the Intensive Care Unit IV Medical Information Market Database between 2008 and 2022; the eICU-CRD cohort consists of patients selected from the eICU Collaborative Research Database between 2014 and 2015.

[0013] In step S2, the candidate predictive factors include demographic parameters, clinical parameters, treatment parameters, laboratory parameters, HIV-specific indicators, and comorbidities. The demographic parameters include age, sex, and length of hospital stay. The clinical parameters include SOFA score, septic shock, and the use of supportive care. The supportive care includes antiretroviral therapy, vasopressors, continuous renal replacement therapy, and mechanical ventilation. The laboratory parameters include albumin, aspartate aminotransferase, creatinine, glucose, international normalized ratio, lactate, lactate dehydrogenase, lymphocyte count, platelet count, total bilirubin, blood urea nitrogen, and white blood cell count. The HIV-specific indicators include CD4+ T cell count and HIV viral load. The comorbidities include coronary heart disease, chronic kidney disease, diabetes, hypertension, cirrhosis, chronic obstructive pulmonary disease, and opportunistic infections.

[0014] The 10-fold cross-validation randomly divides the training data into 10 subsets. Each time, 9 subsets are used to train the prognostic evaluation model, and 1 subset is used to validate the performance of the prognostic evaluation model. This process is repeated 10 times and the average result is taken.

[0015] Step S3 specifically includes the following steps:

[0016] S31: Calculate the prognostic risk score for each sample in the derivation queue, internal validation queue, and external validation queue, and divide the samples into different risk groups based on the prognostic risk score;

[0017] The risk groups are divided as follows: those with a prognostic risk score less than or equal to the first threshold are in the low-risk group; those with a score greater than the first threshold and less than or equal to the second threshold are in the medium-risk group; and those with a score greater than the second threshold and less than the highest theoretical value are in the high-risk group.

[0018] S32: Harrell's C index was used to assess the discriminative power of the prognostic assessment model. The calibration plot was used to compare predicted risk with observed risk to assess the calibration power of the prognostic assessment model. The Akaike information criterion, homogeneity χ², likelihood ratio χ², and Hosmer-Lemeshow χ² test were used to assess the goodness of fit of the prognostic assessment model.

[0019] S33: Decision curve analysis was used to evaluate the clinical utility of the prognostic assessment model, and Kaplan-Meier survival curves and log-rank tests were used to compare survival differences among different risk groups.

[0020] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0021] 1. In the prior art, "CD4+ count is considered the gold standard for HIV prognosis". The prognostic assessment model of this invention incorporates HIV-specific virological parameters (HIV viral load) and for the first time found that in patients with HIV-related sepsis, HIV viral load is a better short-term prognostic predictor than CD4+ T cell count.

[0022] 2. This invention establishes for the first time a prognostic assessment model specifically for HIV-infected sepsis patients, filling a technological gap in this field; it comprehensively covers multiple factors of HIV-related sepsis, and the prognostic assessment model is easy to operate, requiring only routine clinical and laboratory parameters to complete risk stratification, and can be developed into an online calculation tool, suitable for clinical scenarios with limited resources. Attached Figure Description

[0023] Figure 1 This is a flowchart of the method for constructing a prognostic assessment model for HIV-infected sepsis as described in this invention.

[0024] Figure 2 The flowchart for the selection process in the research queue.

[0025] Figure 3 This is a schematic diagram of the risk stratification survival curves for each cohort in HISPS.

[0026] Figure 4 This diagram illustrates the discriminative power and clinical efficacy of the HISPS and SOFA scores. Detailed Implementation

[0027] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0028] like Figure 1 A method for constructing a prognostic assessment model for HIV-infected sepsis includes the following steps:

[0029] S1. Constructing the dataset: Obtain sample data and corresponding clinical information of HIV-infected sepsis patients, and construct the derivation queue, internal validation queue and external validation queue;

[0030] S2. Prognostic Assessment Model Construction: Based on the derivation queue, candidate predictors are obtained through screening using minimum absolute contraction and the Cox regression selection operator, followed by 10-fold cross-validation. Candidate predictors are then eliminated using recursive features to identify key predictors. After classifying key predictors according to clinical relevance thresholds, they are input into a multivariate Cox proportional hazards model, which outputs independent predictors and their corresponding regression coefficients. Each regression coefficient is multiplied by 10 and rounded to the nearest integer to obtain a risk score for each independent predictor. The risk scores of the independent predictors are then summed to obtain the prognostic risk score.

[0031] The independent predictors included age, ongoing renal replacement therapy, septic shock, HIV viral load, international normalized ratio, lactate dehydrogenase, lactate, blood urea nitrogen, and albumin.

[0032] S3. Model Evaluation: The prognostic assessment model was evaluated for discrimination, calibration and clinical utility using the derivation cohort, internal validation cohort and external validation cohort respectively.

[0033] like Figure 2 In step S1, data of patients with sepsis were collected and screened to obtain a deduced cohort, an internal validation cohort, and an external validation cohort. The screening criteria for the sample data were: patients were adults who were diagnosed with HIV infection during hospitalization (by serological or RNA testing) and met the sepsis-3 criteria. The exclusion criteria for patients who did not meet the criteria included: (1) missing key laboratory data; (2) age <18 years; (3) hospitalization time <24 hours; (4) pregnancy; (5) laboratory testing not completed within 48 hours after the onset of sepsis. A total of 2,247 HIV-related sepsis patients were included in the deduced cohort, internal validation cohort, and external validation cohort, of which:

[0034] The derivation cohort and the internal validation cohort were drawn from the hospital’s electronic health record system. The derivation cohort consisted of 1,665 patients admitted between January 2013 and December 2023, while the internal validation cohort consisted of 180 patients admitted between January 2024 and March 2025.

[0035] The external validation cohort consisted of 402 patients, comprising two independent cohorts: the MIMIC-IV cohort and the eICU-CRD cohort. The MIMIC-IV cohort included 240 patients selected from the Critical Care IV Medical Information Marketplace Database (MIMIC-IV, version 3.0) between 2008 and 2022; the eICU-CRD cohort included 162 patients selected from the eICU Collaborative Research Database (eICU-CRD) between 2014 and 2015. MIMIC-IV is a longitudinal, single-center database that includes 94,458 ICU admissions of 65,366 different adults admitted to Boston and Massachusetts between 2008 and 2022. eICU-CRD is a multi-center database developed by Philips Healthcare that covers 208 ICUs in the United States between 2014 and 2015.

[0036] The in-hospital mortality rates for the deduced cohort, internal validation cohort, MIMIC-IV cohort, and eICU-CRD cohort were 22.3%, 28.3%, 12.9%, and 18.5%, respectively (P<0.001), with median ages of 43.9, 51.1, 53.0, and 49.5 years, respectively. In the deduced cohort, most patients had significant immunosuppression, with 89.3% having a CD4 count <200 cells / μL; in the internal validation cohort, 70.6% of patients had a CD4 count <200 cells / μL; and in the external validation cohort, 37.5% of patients in the MIMIC-IV cohort had a CD4 count <200 cells / μL, while CD4 data were unavailable in the eICU-CRD cohort. In the deduced and internal validation cohorts, 56.5% and 41.1% of patients, respectively, had an HIV viral load ≥10. 5 Copy / mL; This variable (viral load data) was not included in the external validation analysis of the final model because more than 50% of patients in the external cohort lacked viral load data.

[0037] Baseline characteristics, clinical manifestations, and laboratory findings of the derivation cohort, internal validation cohort, and external validation cohort were summarized. Missing data were processed using multiple imputation. For laboratory parameter variables with a missing rate of <30%, multiple imputation was performed using the mice package in R software.

[0038] In step S2, among the 1665 patients in the derivation cohort with complete outcome data, the LASSO (Laminate Absolute Contraction and Selection) Cox regression (glmnet package) was used for screening, and 10-fold cross-validation was used to reduce dimensionality to obtain candidate predictors. The 10-fold cross-validation randomly divided the training data into 10 subsets. Each time, 9 subsets were used to train the prognostic assessment model, and 1 subset was used to validate the model's performance. This process was repeated 10 times, and the average result was taken to reduce the risk of overfitting and improve generalization ability. The candidate predictors included demographic factors (age, sex, length of hospital stay), clinical... Clinical parameters (SOFA score, septic shock), treatment (antiretroviral therapy, vasopressors, continuous renal replacement therapy, mechanical ventilation), and laboratory parameters (albumin, aspartate aminotransferase, creatinine, glucose, international normalized ratio, lactate, lactate dehydrogenase, lymphocyte count, platelet count, total bilirubin, blood urea nitrogen, white blood cell count), as well as HIV-specific indicators (CD4+ T cell count, HIV viral load) and comorbidities (coronary artery disease, chronic kidney disease, diabetes, hypertension, cirrhosis, chronic obstructive pulmonary disease, opportunistic infections (such as *Taraxacum multocida*, *Pneumocystis jirovecii* pneumonia)). For laboratory parameters with a missing rate <30%, multiple substitutions were performed using mouse kits.

[0039] To enhance model simplicity, candidate predictors were eliminated through recursive feature analysis, resulting in 13 key predictor variables covering demographic, HIV-related, laboratory, and treatment variables. These key predictor variables were then categorized according to clinical relevance thresholds and input into a multivariate Cox proportional hazards model, as shown in Table 1. The final output consisted of nine independent predictor variables and their corresponding regression coefficients. These independent predictor variables included: age, continuous renal replacement therapy (CRRT), septic shock, HIV viral load, international normalized ratio (INR), lactate dehydrogenase (LDH), lactate, blood urea nitrogen, and albumin. As shown in Table 1, the specific risk score allocations for independent predictor variables are as follows: septic shock: 14 points; international normalized ratio ≥2.0: 9 points; lactate ≥8 mmol / L: 9 points, 4-7.9 mmol / L: 7 points, 2-3.9 mmol / L: 4 points; lactate dehydrogenase ≥750 U / L: 8 points, 500-749 U / L: 7 points, 250-499 U / L: 3 points; continuous renal replacement therapy: 6 points; blood urea nitrogen 7-13.9 mmol / L: 6 points, ≥14 mmol / L: 5 points; age ≥60 years: 4 points; HIV viral load ≥10... 5 3 points for copies / mL; 3 points for albumin <26g / L.

[0040] In Table 1, the hazard ratios with 95% confidence intervals were derived from a multivariate Cox proportional hazards regression model; LASSO-Cox regression was used for variable selection, and 10-fold cross-validation was performed; age was included as a prior covariate, and its importance as a clinical prognostic factor has been established regardless of the selected status; the regression coefficients were converted into integer risk scores, with higher scores indicating a higher risk of death; the two-sided p-values ​​were calculated using the Wald test.

[0041] Table 1

[0042]

[0043] The prognostic assessment model for in-hospital mortality in patients with HIV-related sepsis (HISPS), also known as the prognostic risk score, is the sum of the risk scores of each independent predictor variable. The risk score of each independent predictor variable is calculated by multiplying its corresponding regression coefficient by 10 and rounding it to the nearest integer.

[0044] HISPS was developed as a 48-hour prognostic score, with the indicative time defined as the time of sepsis onset. Predictive variables are limited to variables measured within 48 hours of sepsis onset or interventions initiated. Risk prediction is performed on patients still in the hospital at the 48-hour milestone.

[0045] In step S3, the prognostic assessment model is evaluated using internal and external validation cohorts. HISPS is assigned based on nine independent predictive variables and their corresponding scores, covering demographics (age ≥60 years = 4 points), organ support (continuous renal replacement therapy = 6 points, septic shock = 14 points), and HIV-specific indicators (viral load ≥10...). 5 The HISPS score was calculated as follows: copies / mL (3 points) and laboratory parameters (INR, lactate dehydrogenase, lactate, blood urea nitrogen, and albumin, scored from 3 to 9 points respectively according to their respective levels). The total score ranged from 0 to 62 points. Based on the distribution of HISPS and its association with in-hospital mortality, two optimal cutoff points were determined, dividing patients into a low-risk group (0–12 points, corresponding to an in-hospital mortality rate of approximately 13.0%), a medium-risk group (13–24 points, corresponding to an in-hospital mortality rate of approximately 25.6%), and a high-risk group (25–62 points, corresponding to an in-hospital mortality rate of approximately 55.5%). As shown in Table 2, HISPS demonstrated good discriminative and calibration abilities in both the derivation and validation cohorts. Compared with the SOFA score, HISPS showed better prognostic accuracy. HISPS also showed better Harrell's C index in the derivation cohort. The C-index reached 0.801 (95% confidence interval: 0.775-0.827), with an internal validation cohort C-index of 0.771 (95% confidence interval: 0.695-0.846), the MIMIC-IV external validation cohort C-index of 0.735 (95% confidence interval: 0.622-0.848), and the eICU-CRD external validation cohort C-index of 0.696 (95% confidence interval: 0.576-0.816), all of which are better than the traditional SOFA score. Compared with SOFA, HISPS has a lower Akaike Information Criterion (AIC) value (4463.3 vs 4679.5), and better homogeneity and likelihood ratio χ² statistic, indicating that the present invention has good discriminative power and universality.

[0046] In the external validation cohorts, although the MIMIC-IV cohort lacked HIV viral load data and the eICU-CRD cohort lacked HIV viral load and LDH data, the discriminative power of HISPS (Harrell's C) was still slightly higher than that of SOFA (MIMIC: 0.735 vs 0.725; eICU: 0.696 vs 0.634). Figure 3The Kaplan-Meier survival curves (risk stratification survival curves) of the HISPS prognostic assessment model in each cohort are shown. As the Kaplan-Meier survival analysis shows, in the derived cohort (A), internal validation cohort (B), MIMIC-IV external validation cohort (C), and eICU-CRD external validation cohort (D), the survival differences among the low-risk group (0–12 points), intermediate-risk group (13–24 points), and high-risk group (25–62 points) were statistically significant (log-rank test P<0.01). The survival probability of patients in the high-risk group was significantly lower than that of the low-risk and intermediate-risk groups at all time points, and the separation trend of the survival curves between the groups became more and more obvious with the extension of time, indicating that the model has a good risk discrimination ability.

[0047] To facilitate clinical application, an interactive web-based calculator based on the Shiny software package was developed (https: / / hisps.shinyapps.io / appp / ).

[0048] like Figure 4 This study assesses the discriminative power, clinical utility, and calibration capability of HISPS across various cohorts. Columns 1, 2, 3, and 4 contain data from the derived cohort, internal validation cohort, MIMIC-IV cohort, and eICU-CRD cohort, respectively. Row A represents Harrell's C index (95% confidence interval), showing that HISPS outperforms the SOFA score in all four cohorts in terms of discriminative power. Row B represents decision curve analysis, demonstrating that HISPS provides superior net clinical benefits compared to SOFA across all cohorts. Row C represents the calibration plot, showing a high degree of consistency between the hospital mortality rate predicted by HISPS and the observed actual hospital mortality rate across all risk groups. The calibration points are generally distributed along a 45° diagonal, indicating good model calibration capability. In summary, the overall assessment demonstrates that HISPS can reliably and accurately assess the prognosis and risk stratification of HIV-infected sepsis patients across various clinical scenarios and databases.

[0049] Table 2

[0050]

[0051] All analyses and evaluations of the prognostic assessment model were performed using R software (version 4.4.0). A two-sided p-value <0.05 was considered statistically significant. Continuous variables included age, SOFA score, and various laboratory indicators (such as albumin, creatinine, lactate, LDH, etc.), expressed as mean ± standard deviation or median (interquartile range), and were analyzed using t-tests or Mann–Whitney U tests according to their distribution. Categorical variables included gender, septic shock, treatment measures (ART, CRRT, mechanical ventilation, vasoactive drug use), and comorbidities, expressed as frequency (percentage), and were compared using χ² tests or Fisher's exact test.

[0052] For example, consider a 45-year-old male HIV-infected septicemia patient diagnosed with septic shock. His INR was 2.3, lactate was 5.2 mmol / L, LDH was 320 U / L, age was 45, and HIV viral load was ≥10. 5 The patient's blood alcohol content was 100 copies / mL, albumin was 25 g / L, urea nitrogen was 8.5 mmol / L, and the patient did not receive CRRT. According to the scoring rules above, the scores for each variable for this patient were as follows: septic shock 14 points, INR ≥ 2.0 9 points, lactate 4-7.9 mmol / L 7 points, LDH 250-499 U / L 3 points, age < 45 years 0 points, HIV viral load ≥ 10 5 The patient's HISPS score is calculated as follows: copies / mL = 3 points, albumin <26g / L = 3 points, blood urea nitrogen 7-13.9mmol / L = 6 points, and CRRT not received = 0 points. The total HISPS score is 45 points. Therefore, this patient belongs to the high-risk group (25-62 points), with a corresponding mortality rate of 55.5%. Based on this, clinicians can implement more intensive monitoring and more aggressive intervention measures for this patient.

[0053] Existing prognostic models are mostly developed based on the general population. Because they do not take into account factors such as HIV-related immunodeficiency and chronic immune activation, their predictive efficacy in HIV-infected individuals is significantly reduced. This invention directly addresses and solves the unique technical challenges of prognostic assessment for HIV-infected sepsis, aiming to build a dedicated model for this specific population. The technical problem it solves is itself different from existing technologies.

[0054] This invention uses LASSO-Cox regression to data-drivenly screen 9 independent predictive factors from 31 candidate variables and assigns them specific weights (i.e., risk scores, which are integer scores used to construct a risk scoring system based on the regression coefficients of each variable in the multivariate Cox regression model after standardization, reflecting the relative contribution of each variable to the risk of death). In this process, this invention discovers for the first time that in patients with HIV-related sepsis, HIV viral load is a superior short-term prognostic predictor compared to the traditional indicator CD4+ T cell count. This finding contradicts the conventional understanding that "CD4+ count is the gold standard for HIV prognosis," demonstrating the inventors' profound understanding of specific pathological mechanisms, and the unpredictable nature of its technical effects. Furthermore, this invention includes a specific combination of predictive factors (age ≥60 years, continuous renal replacement therapy, septic shock, HIV viral load ≥10...). 5 The values ​​(copy / mL, INR, LDH, lactate, urea nitrogen, and albumin) and their weighting are not simply a combination of existing technologies.

[0055] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for constructing a prognostic assessment model for HIV-infected sepsis, characterized in that, Includes the following steps: S1. Constructing the dataset: Obtain sample data and corresponding clinical information of HIV-infected sepsis patients, and construct the derivation queue, internal validation queue and external validation queue; S2. Prognostic Assessment Model Construction: Based on the derivation queue, candidate predictors are obtained through screening using minimum absolute contraction and the Cox regression selection operator, followed by 10-fold cross-validation. Candidate predictors are then eliminated using recursive features to identify key predictors. After classifying key predictors according to clinical relevance thresholds, they are input into a multivariate Cox proportional hazards model, which outputs independent predictors and their corresponding regression coefficients. Each regression coefficient is multiplied by 10 and rounded to the nearest integer to obtain a risk score for each independent predictor. The risk scores of the independent predictors are then summed to obtain the prognostic risk score. The independent predictors included age, ongoing renal replacement therapy, septic shock, HIV viral load, international normalized ratio, lactate dehydrogenase, lactate, blood urea nitrogen, and albumin. S3. Model Evaluation: The prognostic assessment model was evaluated for discrimination, calibration and clinical utility using the derivation cohort, internal validation cohort and external validation cohort respectively.

2. The method for constructing a prognostic assessment model for HIV-infected sepsis according to claim 1, characterized in that, Step S1 involves obtaining sample data and corresponding clinical information from HIV-infected sepsis patients, and screening the sample data. The screening criteria for the sample data are: the patients are adults who are diagnosed with HIV infection during hospitalization and meet the sepsis-3 criteria. Patients who did not meet the criteria were excluded. The exclusion criteria included: missing key laboratory data; age <18 years; length of hospital stay <24 hours; pregnancy; and laboratory tests not completed within 48 hours of the onset of sepsis. Based on the sample data obtained after screening, an inference queue, an internal validation queue, and an external validation queue are constructed. The inference queue and the internal validation queue are from the hospital's electronic health record system. The inference queue consists of patients admitted between January 2013 and December 2023, and the internal validation queue consists of patients admitted between January 2024 and March 2025. The external validation cohort consists of two independent cohorts: the MIMIC-IV cohort and the eICU-CRD cohort. The MIMIC-IV cohort consists of patients selected from the Intensive Care Unit IV Medical Information Market Database between 2008 and 2022; the eICU-CRD cohort consists of patients selected from the eICU Collaborative Research Database between 2014 and 2015.

3. The method for constructing a prognostic assessment model for HIV-infected sepsis according to claim 1, characterized in that, In step S2, the candidate predictive factors include demographic parameters, clinical parameters, treatment parameters, laboratory parameters, HIV-specific indicators, and comorbidities. The demographic parameters include age, sex, and length of hospital stay. The clinical parameters include SOFA score, septic shock, and the use of supportive care. The supportive care includes antiretroviral therapy, vasopressors, continuous renal replacement therapy, and mechanical ventilation. The laboratory parameters include albumin, aspartate aminotransferase, creatinine, glucose, international normalized ratio, lactate, lactate dehydrogenase, lymphocyte count, platelet count, total bilirubin, blood urea nitrogen, and white blood cell count. The HIV-specific indicators include CD4+ T cell count and HIV viral load. The comorbidities include coronary heart disease, chronic kidney disease, diabetes, hypertension, cirrhosis, chronic obstructive pulmonary disease, and opportunistic infections. The 10-fold cross-validation randomly divides the training data into 10 subsets. Each time, 9 subsets are used to train the prognostic evaluation model, and 1 subset is used to validate the performance of the prognostic evaluation model. This process is repeated 10 times and the average result is taken.

4. The method for constructing a prognostic assessment model for HIV-infected sepsis according to claim 1, characterized in that, Step S3 specifically includes the following steps: S31: Calculate the prognostic risk score for each sample in the derivation queue, internal validation queue, and external validation queue, and divide the samples into different risk groups based on the prognostic risk score; The risk groups are divided as follows: those with a prognostic risk score less than or equal to the first threshold are in the low-risk group; those with a score greater than the first threshold and less than or equal to the second threshold are in the medium-risk group; and those with a score greater than the second threshold and less than the highest theoretical value are in the high-risk group. S32: Harrell's C index was used to assess the discriminative power of the prognostic assessment model. The calibration plot was used to compare predicted risk with observed risk to assess the calibration power of the prognostic assessment model. The Akaike information criterion, homogeneity χ², likelihood ratio χ², and Hosmer-Lemeshow χ² test were used to assess the goodness of fit of the prognostic assessment model. S33: Decision curve analysis was used to evaluate the clinical utility of the prognostic assessment model, and Kaplan-Meier survival curves and log-rank tests were used to compare survival differences among different risk groups.

Images