A method, system, electronic device, and storage medium for assessing the risk of in-hospital mortality associated with AIDS-related Basilella marneffei infection.

Abstract

This invention discloses a method, system, electronic device, and storage medium for assessing the risk of in-hospital mortality in AIDS-related Marneffei infection. The method includes the following steps: obtaining test values ​​for four indicators: ALB, PCT, BUN, and WBC; standardizing the units of the four indicator test values; generating corresponding variables A, P, U, and B from the standardized test values ​​according to preset thresholds; calculating the linear predictive value z and the in-hospital mortality prediction probability Pr using the A-PUB scoring model; and comparing the in-hospital mortality prediction probability with a preset risk level classification threshold to obtain the corresponding risk level. This method and system use only four routine test indicators from the early stages of patient admission to quickly output risk before the pathogen culture results are available, avoiding reliance on complex inflammation ratios and non-widely available tests. It is convenient for application in emergency departments, infectious disease departments, and primary hospitals, reduces manual calculation errors, and shortens the risk assessment time.

Description

Technical Field

[0001] This invention relates to the field of medical data processing and clinical risk assessment technology, specifically to an in-hospital mortality risk assessment system, electronic equipment, and storage medium for AIDS-related malonfibrillaria disease. In particular, it relates to an A-PUB scoring model constructed using four routine admission test values, as well as a system, electronic equipment, and computer-readable storage medium for implementing the model's data acquisition, standardization, probability calculation, risk stratification, interpretation, and early warning output. Background Technology

[0002] Marneffei infection is an invasive opportunistic fungal infection caused by *Talaromyces marneffei* (TM), and is particularly common in patients with advanced acquired immunodeficiency syndrome (AIDS) in southern my country and Southeast Asia. The disease can affect the lungs, blood, bone marrow, lymph nodes, skin, and abdominal organs. Common clinical manifestations include fever, anemia, cough, weight loss, lymphadenopathy, and multi-organ involvement, but these manifestations lack specificity.

[0003] Pathogen culture remains a crucial basis for diagnosis, but etiological diagnosis relies on the culture or pathological examination of specimens such as blood, bone marrow, lymph nodes, skin, and bronchoalveolar lavage fluid, resulting in a long turnaround time. The median time from admission to diagnosis and positive blood culture is approximately 6 days, and some patients experience organ dysfunction or death before diagnosis. Clinically, short-term risk is often assessed based on CD4+ T cell count, symptom severity, individual inflammatory markers, or physician experience; however, in patients with generally extremely low CD4+ T cell levels, CD4+ T cell count has limited ability to differentiate in-hospital mortality. Previous prognostic models have numerous variables, including complex inflammatory ratios, non-widely used tests, or post-treatment variables, making them difficult to use immediately in emergency and primary care hospitals. Some models only report statistically relevant factors or simple total scores, without providing directly deployable probability formulas, automatic unit conversion, stratification thresholds, or interpretable outputs. Using treatment or process variables such as antifungal treatment, length of hospital stay, and time to diagnosis for admission prediction can easily lead to time-dependent bias and information leakage, hindering real-world deployment.

[0004] Therefore, the existing technology has the following shortcomings in the diagnosis of malonfibrillary scabies: 1) Insufficient timeliness: It cannot quickly identify high-risk patients before the culture results are available; 2) Insufficient accessibility: It relies on special tests, complex parameters or manual calculations, making it difficult to integrate into hospital information systems; 3) Insufficient standardization: The input methods and missing value handling of different units and systems are inconsistent, which can easily lead to calculation errors; 4) Insufficient interpretability: The model output cannot clearly show which indicators drive the risk increase, resulting in limited clinical acceptance; 5) Insufficient transferability: It lacks model version management, external calibration and threshold configuration mechanisms; 6) Insufficient specificity: There is still a lack of a concise probabilistic model for AIDS-related malonfibrillary scabies based on the values ​​of the four routine admission tests (albumin (ALB), procalcitonin (PCT), blood urea nitrogen (BUN) and white blood cell count (WBC)).

[0005] Therefore, there is an urgent clinical need for an automated tool that can quantify and stratify mortality risk before etiological diagnosis by utilizing routine early admission test results. Summary of the Invention

[0006] This invention aims to provide a method, system, electronic device, and storage medium for assessing the risk of in-hospital mortality in patients with AIDS-related malonfibrillaria disease. The system automatically extracts, verifies, and standardizes four routine admission test values ​​that are scattered and inconsistent between units in the hospital's Laboratory Information System (LIS) and Electronic Medical Record System (EMR). It then generates in-hospital mortality risk, risk level, and early warning information in real time through an interpretable probability model. This reduces errors from manual calculations, shortens risk assessment time, and provides standardized data support for triage and monitoring prioritization before etiological diagnosis.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] In a first aspect, the present invention provides a method for assessing the risk of in-hospital mortality from AIDS-related Marneffei infection based on the A-PUB scoring model, comprising the following steps:

[0009] S1. Obtain the test values ​​of four indicators in the early stage of patient admission: ALB, PCT, BUN and WBC.

[0010] S2. Standardize and convert the units of the test values ​​for the four indicators;

[0011] S3. Generate corresponding variables A, P, U, and B from the standardized test values ​​according to the preset binarization threshold;

[0012] S4. Calculate the linear prediction value z and the in-hospital mortality prediction probability Pr using the A-PUB scoring model. The formula for calculating the linear prediction value is z = -4.598 + 2.022 × A + 1.166 × P + 1.189 × U + 1.467 × B. The formula for calculating the in-hospital mortality prediction probability is Pr = 1 / (1 + e^{-z}), where e is a natural constant.

[0013] S5. Compare the predicted probability of in-hospital mortality with the preset risk level classification threshold to obtain the corresponding risk level.

[0014] Preferably, the standard units for the four indicators ALB, PCT, BUN and WBC are g / L, ng / mL, mmol / L and ×10^9 / L, respectively.

[0015] Preferably, the calculation formulas for A, P, U, and B are A=I (ALB<22 g / L), P=I (PCT≥1.0 ng / mL), U=I (BUN≥6.0 mmol / L), and B=I (WBC≥5.0×10^9 / L), respectively, where I(·) is an indicator function, which is assigned a value of 1 when the condition in parentheses is true, and a value of 0 otherwise.

[0016] Preferably, the risk levels include: a low-risk level with Pr < 0.05, a medium-risk level with 0.05 ≤ Pr < 0.20, and a high-risk level with Pr ≥ 0.20.

[0017] Preferably, step S5 further includes comparing the predicted probability of in-hospital death with a preset early warning threshold to obtain a corresponding early warning indicator, wherein the early warning indicator includes: a yellow warning with Pr≥0.10 and a red high-risk warning with Pr≥0.20.

[0018] Secondly, the present invention provides an AIDS-related Malneffei infection in-hospital mortality risk assessment system based on the A-PUB scoring model, including a data acquisition module, a data standardization module, a threshold coding module, a probability calculation module, and a risk stratification and early warning output module;

[0019] The data acquisition module is used to obtain the test values ​​of four patient indicators: ALB, PCT, BUN, and WBC.

[0020] The data standardization module is used to convert the original units of the acquired test values ​​into preset standardized units.

[0021] The threshold encoding module is used to generate corresponding variables A, P, U, and B from the standardized test values ​​according to four preset thresholds based on binarization rules.

[0022] The probability calculation module is used to call variables A, P, U, and B to calculate the linear prediction value z and the in-hospital mortality prediction probability Pr through the A-PUB scoring model. The formula for calculating the linear prediction value is z = -4.598 + 2.022 × A + 1.166 × P + 1.189 × U + 1.467 × B; the formula for calculating the in-hospital mortality prediction probability is Pr = 1 / (1 + e^{-z}), where e is a natural constant.

[0023] The risk stratification and early warning output module is used to compare the in-hospital mortality prediction probability Pr with the preset risk stratification threshold and the preset early warning threshold, and output the corresponding risk stratification and early warning indicator.

[0024] Preferably, the four patient indicators are obtained from LIS, EMR or manually entered. The data standardization module also includes time window filtering, duplicate value removal, reasonable range check and missing value prompting for the obtained test values. The standard units of the four indicators ALB, PCT, BUN and WBC are g / L, ng / mL, mmol / L and ×10^9 / L, respectively.

[0025] Preferably, the calculation formulas for A, P, U, and B are A=I (ALB<22 g / L), P=I (PCT≥1.0 ng / mL), U=I (BUN≥6.0 mmol / L), and B=I (WBC≥5.0×10^9 / L), respectively, where I(·) is an indicator function. When the condition in parentheses is true, the function is assigned a value of 1, otherwise it is assigned a value of 0.

[0026] Preferably, the risk levels include: a low-risk level with Pr < 0.05, a medium-risk level with 0.05 ≤ Pr < 0.20, and a high-risk level with Pr ≥ 0.20.

[0027] Preferably, the early warning indicators include: a yellow warning with Pr ≥ 0.10 and a red high-risk warning with Pr ≥ 0.20.

[0028] Preferably, the system further includes an explanation module, a display and push module, and a version and audit module; the explanation module is used to output the contribution of each variable to the prediction result; the display and push module is used to display the result on the doctor's workstation, ward screen, webpage, mobile terminal or report; and the version and audit module is used to manage model version, threshold configuration, external calibration parameters and complete calculation log.

[0029] Thirdly, the present invention provides an electronic device including at least one memory, a processor, a communication interface, a display or output interface, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the in-hospital mortality risk assessment method described above.

[0030] Fourthly, the present invention provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the in-hospital mortality risk assessment method described above.

[0031] Preferably, the computer-readable storage medium may be a disk, solid-state storage, server storage, cloud object storage, or a non-transitory computer-readable medium.

[0032] Compared with existing technologies, the beneficial effects of this solution are:

[0033] The method and system for assessing the risk of in-hospital mortality in AIDS-related *Basilella marneffei* disease based on the A-PUB scoring model provided by this invention have outstanding advantages in terms of early detection and simplicity. Using only four routine laboratory indicators from the early admission stage, it can quickly output risk before the pathogen culture results are available, avoiding reliance on complex inflammation ratios and non-widely available tests, making it convenient for application in emergency departments, infectious disease departments, and primary care hospitals. In terms of probabilistics, the A-PUB scoring model retains the regression weights of different variables, directly outputting individualized probabilities rather than a coarse total score. Its standardized design includes built-in unit conversion, time window selection, missing value indication, and outlier verification, reducing errors from manual calculations. Furthermore, the model is interpretable, capable of displaying the risk direction and contribution of each variable, and can generate SHAP beehive plots or waterfall plots. It also supports embedding in LIS, EMR, clinical workstations, web pages, mobile devices, or laboratory equipment, and has configurable capabilities for threshold adjustment, model version management, and external cohort recalibration. Clinical efficacy assessments showed that the model's internal validation AUC, Brier score, cross-validation, and decision curve results all indicated good discrimination, calibration, and potential net benefit, providing a practical tool for early risk stratification and clinical decision-making. Furthermore, the output is positioned as an auxiliary risk assessment tool, not a direct replacement for etiological diagnosis and clinical treatment decisions, and retains manual review and auditing traces, thus possessing strong technical security. Attached Figure Description

[0034] Figure 1 A schematic diagram of the A-PUB scoring model calculation and early warning process;

[0035] Figure 2 Example of a SHAP bee colony graph for the A-PUB rating model;

[0036] Figure 3 Example of a SHAP waterfall chart for the A-PUB rating model;

[0037] Figure 4 A graph comparing the A-PUB scoring model with the AUC of individual indicators;

[0038] Figure 5Comparison of the A-PUB rating model with the ROC curve of individual indicators;

[0039] Figure 6 A calibration chart for the A-PUB scoring model;

[0040] Figure 7 To predict risk stratification and observe in-hospital mortality using the A-PUB scoring model;

[0041] Figure 8 This is a schematic diagram of the overall architecture of an AIDS-related Marneffei basket disease in-hospital mortality risk assessment system based on the A-PUB scoring model;

[0042] Figure 9 An electronic device framework diagram for implementing the in-hospital mortality risk assessment method or system provided by this invention. Detailed Implementation

[0043] To facilitate understanding of the present invention by those skilled in the art, the technical solution of the present invention will be further described in detail below with reference to embodiments and accompanying drawings. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0044] In this invention, unless otherwise stated, scientific and technical terms used herein have the meanings commonly understood by those skilled in the art.

[0045] In this invention, the term "A-PUB" corresponds to Albumin (ALB), Procalcitonin (PCT), Urea nitrogen (BUN), and Blood leukocyte count (WBC), respectively.

[0046] In this invention, the “A-PUB scoring model” is not a simple equal-weighted scoring system, but rather retains the Logistic regression weights of the four binary classification indicators to output an individualized in-hospital mortality prediction probability.

[0047] The establishment of the A-PUB scoring model in this invention will be described below.

[0048] First, the data (cohort) for establishing the A-PUB scoring model of this invention came from electronic medical records, laboratory information systems, imaging records, and microbiological / pathological examination records of the Department of Infectious Diseases and related inpatient wards of the First Affiliated Hospital of Fujian Medical University in Quanzhou from October 2014 to September 2022. A total of 133 adult patients diagnosed with AIDS complicated with TM infection during hospitalization were included, of whom 103 were discharged after improvement, 13 were discharged voluntarily after deterioration, and 17 died in hospital.

[0049] The diagnosis of AIDS was based on a positive initial HIV antibody screening at our hospital, confirmed by the Quanzhou Municipal Center for Disease Control and Prevention. The diagnosis of TM infection was primarily based on the isolation or detection of TM in blood, bone marrow, lymph nodes, skin, bronchoalveolar lavage fluid, pleural effusion, peritoneal effusion, or other clinical specimens; patients with epidemiological exposure, typical clinical and imaging manifestations, and supporting etiological or pathological evidence were also included. Patients under 18 years of age, lacking clear etiological evidence, with incomplete key clinical data, or who had received systemic antifungal treatment prior to admission were excluded.

[0050] The difference between this invention and the prior art is that the prior art mainly describes the general clinical characteristics of the cohort and the factors related to the improvement / deterioration of the outcome; while this invention redefines the technical objective as "calculation of the probability of in-hospital mortality in the early stage of admission", using only information available in the early stage of admission, excluding process variables such as length of hospital stay, time of diagnosis, time of positive blood culture, post-treatment indicators and antifungal treatment, so as to avoid leakage of outcome information.

[0051] Secondly, the process of establishing the A-PUB scoring model and selecting variables in this invention includes the following:

[0052] 1) Univariate analysis of all variables: The inclusion criteria cover demographic, symptom and sign, comorbidities / opportunistic infections, treatment description, laboratory, immunological, mycological, radiological and etiological variables; in the main text, candidate variables with P<0.05 are retained and OR, 95% CI and P value are recorded.

[0053] 2) LASSO dimensionality reduction: After excluding post-treatment variables, significant outcome leakage variables, and highly missing variables, LASSO logistic regression was performed on the admission candidate variables; strict LASSO prioritized the retention of ALB, BUN, and WBC.

[0054] 3) Stable selection: The frequency of each variable entering the regularized model is evaluated by Bootstrap repeated sampling, reducing the randomness caused by a single sampling.

[0055] 4) Elastic network sensitivity analysis: Considering the correlation between inflammation and organ stress variables, elastic network regression was used to verify variable stability; PCT may be compressed due to correlation in strict LASSO, but its single-index AUC is high and it is included in the model in elastic network analysis.

[0056] 5) Multivariate Logistic Regression: Based on the principles of comprehensive statistical signal, clinical interpretability, bedside accessibility, model simplicity, and avoiding collinear redundancy, the four binary variables ALB, PCT, BUN, and WBC are ultimately retained.

[0057] Specifically, the establishment and internal validation of the A-PUB scoring model of this invention includes:

[0058] Data from 133 patients meeting the inclusion criteria were extracted from the hospital information system, with in-hospital mortality as the primary outcome. A univariate analysis of all variables was first performed on demographic, symptom and sign, comorbidities, opportunistic infections, laboratory, immunological, mycological, imaging, and etiological variables. For candidate admission variables used for real-time prediction, after excluding post-treatment and outcome leakage variables, LASSO and Bootstrap stability selection were performed, followed by elasticity network sensitivity analysis to process relevant variables. Based on statistical signals and clinical usability, four indicators—ALB, PCT, BUN, and WBC—were determined and binarized according to the thresholds in Table 1 to establish a multivariate logistic model. Performance was evaluated using ROC, Brier score, calibration, decision curve analysis, cross-validation, and Bootstrap optimistic correction. The results are shown in Table 2.

[0059] Table 1 Input Data and Data Dictionary

[0060] code index Standard units Binarization rules Clinical / Technical Implications A Albumin ALB g / L <22 g / L is recorded as 1, otherwise 0. Reflects nutrient reserves, liver synthesis function, and inflammation / capillary leakage. P Procalcitonin PCT ng / mL ≥1.0 ng / mL is recorded as 1, otherwise 0. Reflects systemic inflammatory load and severe infection response U Blood urea nitrogen (BUN) mmol / L ≥6.0 mmol / L is recorded as 1, otherwise 0. Reflects renal perfusion, catabolism, dehydration, and organ stress. B White blood cell count (WBC) ×10^9 / L ≥5.0×10^9 / L is recorded as 1, otherwise 0. Relative leukocyte response in the context of late immunosuppression

[0061] The formula for calculating the linear prediction value of the multifactor Logistic model (A-PUB rating model) established based on the above is as follows:

[0062] z=-4.598+2.022×A+1.166×P+1.189×U+1.467×B;

[0063] Among them, A, P, U, and B are 0 / 1 variables representing whether the threshold conditions of the four indicators are met, respectively;

[0064] The calculation formulas for A, P, U, and B are A=I(ALB<22 g / L), P=I(PCT≥1.0 ng / mL), U=I(BUN≥6.0 mmol / L), and B=I(WBC≥5.0×10^9 / L), respectively. I(·) is an indicator function. When the condition in parentheses is true, the function is assigned a value of 1, otherwise it is assigned a value of 0.

[0065] The formula for calculating the probability of in-hospital mortality is as follows:

[0066] Pr = 1 / (1+e^{-z}), where e is the natural constant.

[0067] Table 2 Results of multivariate logistic regression analysis

[0068] item condition regression coefficient β OR 95% CI intercept — -4.598 — — A WHITE<22 g / L 2.022 7.55 1.82-31.38 P PCT ≥ 1.0 ng / mL 1.166 3.21 0.70-14.81 U BUN ≥ 6.0 mmol / L 1.189 3.28 0.83-12.91 B WBC ≥ 5.0 × 10^9 / L 1.467 4.34 1.16-16.18

[0069] Although this model uses the name "A-PUB scoring model," it is a weighted probability model and does not replace coefficients with a simple total score. This system can simultaneously display binary states, the contribution of each coefficient, the linear predicted value z, and the final probability Pr. When deployed to an external queue, only the intercept or all coefficients can be recalibrated while retaining the definitions of the four indicators and the risk direction, and the model version number can be saved.

[0070] Figure 1 This is a schematic diagram of the A-PUB scoring model calculation and early warning process (flowchart of in-hospital mortality risk assessment method). The following section further describes the AIDS-related Marneffei basket disease in-hospital mortality risk assessment method based on the A-PUB scoring model.

[0071] The in-hospital mortality risk assessment method provided by this invention is based on the A-PUB scoring model and first obtains four indicators of the patient: ALB, PCT, BUN, WBC and admission time.

[0072] Then, the effective values ​​of the four indicators closest to the admission time are selected from the patient's first test results after admission or within a preset time window (e.g., within 24 hours of admission). If the original units of the selected four indicators are different from the preset standardized units, the units of the four indicators are converted. For example, when converting ALB from g / dL to g / L, the units are multiplied by 10; the values ​​of PCT in μg / L and ng / mL are equivalent; the values ​​of BUN from mg / dL to mmol / L are divided by 2.801; and the values ​​of WBC in 10^3 / μL and 10^9 / L are equivalent. If any of the four indicators is missing, the preferred implementation is to prompt "data incomplete" and request a supplementary test, without directly outputting the final clinical risk.

[0073] Subsequently, the standardized test values ​​were used to generate corresponding variables A, P, U, and B according to the four pre-set thresholds of the binarization rules.

[0074] Next, the linear predicted value z and the in-hospital mortality predicted probability Pr were calculated using a multifactor logistic model.

[0075] Finally, based on the predicted probabilities, the following risk levels are output: Pr < 0.05 indicates low risk; 0.05 ≤ Pr < 0.20 indicates medium risk; and Pr ≥ 0.20 indicates high risk. To improve the sensitivity of early warnings, a yellow warning threshold of Pr ≥ 0.10 can be set; a red high-risk warning is triggered when Pr ≥ 0.20. All thresholds can be configured by hospital administrators based on external validation results.

[0076] It should be noted that the in-hospital mortality risk assessment method provided by this invention can generate global or local interpretations for each patient. The global interpretation is used to show the overall contribution distribution of the four variables to the model output, such as... Figure 2 As shown, Figure 3 Each point in the graph represents a patient, the horizontal axis represents the contribution of a variable to the model output, and the color indicates the magnitude of the variable value. Local interpretation describes the contribution of a high-risk patient from the model's baseline risk to the final predicted risk, such as... Figure 3 As shown. The SHAP value or equivalent contribution algorithm is only used as an explanatory layer, does not change the model variables and coefficients, and is not used for causal inference.

[0077] It should also be noted that all inputs to the A-PUB scoring model of this invention come from routine hospital tests, which are usually available early in the hospital admission process. It does not rely on culture results or high-cost omics testing, thus exhibiting high feasibility. The model development employed univariate analysis, LASSO dimensionality reduction, Bootstrap stability selection, elasticity network sensitivity analysis, and multivariate logistic regression. ROC, calibration, thresholding, risk stratification, decision curve analysis, repeated stratified 5-fold cross-validation, and Bootstrap optimistic correction were also performed (see Table 3 and...). Figure 4 , 5 6 and 7).

[0078] Table 3 Validation results of the A-PUB scoring model

[0079] index result explain A-PUB rating model AUC 0.891 95% bootstrap CI 0.782-0.966 Brier rating 0.070 Overall prediction error is low Repeated stratified 5-fold cross-validation AUC 0.866 Internal resampling validation Bootstrap optimistic correction AUC 0.877 Correcting apparent optimism bias Predicted probability threshold of 0.10 Sensitivity: 88.2%; Negative predictive value: 97.9% Suitable as an early warning Predicted probability threshold of 0.20 Specificity: 91.4%; Positive predictive value: 52.4% Suitable for high-risk triage

[0080] The results showed that the AUC values ​​of individual indicators ALB, PCT, BUN, WBC, and CD4+ T cell count were 0.830, 0.813, 0.781, 0.763, and 0.585, respectively, while the AUC of the A-PUB scoring model was 0.891, indicating that the combined weighting of the four conventional indicators can integrate complementary information. The above data are used to illustrate the technical effectiveness and do not represent independent external validation. Furthermore, based on the in-hospital mortality prediction probability output by the A-PUB scoring model, the study population was stratified by risk. The actual in-hospital mortality rates for the low-risk (<0.05), medium-risk (0.05-<0.20), and high-risk (≥0.20) groups were 1.2% (1 / 83), 17.2% (5 / 29), and 52.4% (11 / 21), respectively. The actual mortality rate increased significantly with increasing predicted risk level, confirming that the model has good discriminative power. The calibration curves show that the observation points of each group are basically distributed along the ideal standard diagonal. The predicted probability of the model is in good agreement with the actual occurrence rate of the outcome. Only a few groups show slight deviations, indicating that the model is well calibrated.

[0081] Figure 8This is a schematic diagram of the overall architecture of the AIDS-related Marneffei basket disease in-hospital mortality risk assessment system based on the A-PUB scoring model of the present invention. The system includes: a data acquisition module, a data standardization module, a threshold coding module, a probability calculation module, and a risk stratification and early warning output module.

[0082] The data acquisition module is used to collect the test values ​​of four indicators of the patient: ALB, PCT, BUN, and WBC, as well as the admission time.

[0083] The data standardization module is used to perform time window filtering, duplicate value removal, reasonableness range checking, and missing value prompting on the acquired test values, and to convert the original units of the test values ​​in the early stage of patient admission into preset standardized units.

[0084] The threshold encoding module is used to generate corresponding variables A, P, U, and B from the standardized test values ​​according to four preset thresholds based on the binarization rules.

[0085] The probability calculation module is used to call the model coefficients to calculate the linear predicted value z and the in-hospital mortality prediction probability Pr through the weighted logistic model. The model coefficients are variables A, P, U, and B generated by the threshold coding module, and the weighted logistic model is the A-PUB scoring model.

[0086] The risk stratification and early warning output module is used to compare the in-hospital mortality prediction probability Pr with the preset risk stratification threshold and the preset early warning threshold, and output the corresponding low, medium or high risk stratification and yellow or red early warning signs.

[0087] In addition, the system includes an interpretation module for outputting the contribution of each variable to the prediction results, wherein the contribution adopts regression coefficient contribution, SHAP value or equivalent interpretation algorithm; the interpretation is only used to explain the weighted logistic model and is not used to re-select variables; a display and push module for displaying the results on doctor workstations, ward screens, web pages, mobile devices or reports; and a version and audit module for managing model version, threshold configuration, external calibration parameters and complete calculation logs.

[0088] Based on the above modules, the system of the present invention can output the in-hospital mortality prediction probability, risk level, main risk contribution and early warning information by inputting the test values ​​of ALB, PCT, BUN and WBC in the early stage of patient admission, thereby reducing the error of manual calculation and shortening the risk assessment time, and providing standardized data support for triage and monitoring priority before etiological diagnosis.

[0089] The following section further describes the methodological procedures executed by each module of the AIDS-related Malneffei infection in-hospital mortality risk assessment system based on the A-PUB scoring model of this invention.

[0090] First, the data acquisition module obtains the patient's four indicators—ALB, PCT, BUN, WBC—and admission time from LIS and EMR via a database interface or through manual entry.

[0091] Then, the data standardization module selects the valid values ​​of four indicators closest to the admission time from the patient's first test results after admission or within a preset time window (e.g., within 24 hours of admission). If the original units of the selected four indicators are different from the preset standardized units, the units of the four indicators are converted.

[0092] If any of the four indicators is missing, the preferred implementation is to display a "data incomplete" message and request supplementary testing, without directly outputting the final clinical risk. It should be noted that in research mode, the system can be configured with validated imputation strategies, but the imputation results should be clearly indicated in the report. The system should also perform timestamp verification, duplicate value handling, anomaly range alerts, and unit consistency checks.

[0093] Subsequently, the threshold encoding module generates corresponding variables A, P, U, and B from the standardized test values ​​according to the four preset thresholds of the binarization rules.

[0094] Next, the linear prediction value z and the in-hospital mortality prediction probability Pr are calculated using the probability calculation module.

[0095] Finally, the risk stratification and early warning output module of this system outputs the following risk levels based on predicted probabilities: Pr < 0.05 indicates low risk; 0.05 ≤ Pr < 0.20 indicates medium risk; and Pr ≥ 0.20 indicates high risk. To improve the sensitivity of early warnings, a yellow warning threshold of Pr ≥ 0.10 can be set; a red high-risk warning is triggered when Pr ≥ 0.20. All thresholds can be configured by hospital administrators based on external verification results, but the system should record the version, the person who modified it, and the effective time. The output items of this system are shown in Table 4.

[0096] Table 4 Output Project Information

[0097] Output items content Predicted probability Display of values ​​and percentages between 0 and 1 Risk level Low risk / medium risk / high risk, configurable color indicators Warning information Push to the clinical workstation when the threshold of 0.10 or 0.20 is reached. Factor Explanation Display whether ALB, PCT, BUN, and WBC exceed the threshold and their model contributions. Data quality Displays the detection time, original units, converted units, and missing / abnormal information. Audit Records Record the model version, computation time, operator, input values, and output results.

[0098] The system of this invention can be deployed on hospital intranet servers, independent workstations, cloud platforms, mobile terminals, or embedded laboratory analysis equipment. A preferred solution adopts a front-end / back-end separation architecture: the back-end service is responsible for model calculation, authorization, and auditing; the front-end is responsible for result display and manual review. To protect patient information, role-based access control, transmission encryption, de-identification, access logging, and result tracking can be implemented.

[0099] Based on the same inventive concept, such as Figure 9As shown, the present invention also provides an electronic device for running the in-hospital mortality risk assessment method or system provided by the present invention. This electronic device includes at least one processor 101, a memory 102, a communication interface 103, and a display / output interface 104, as well as a computer program stored in the memory 102 and executable on the processor 101. When the processor executes the program, it completes data acquisition, unit standardization, threshold encoding, logistic probability calculation, risk stratification, interpretation output, and log recording. The processor 101, communication interface 103, memory 102, and display / output interface 104 communicate with each other via a bus. The communication interface serves as a bridge for data exchange between the electronic device and other devices or modules, enabling electrical connections, protocol conversion, and data format matching between different devices. The display / output interface is used to present the results processed by the processor to the user or external devices in a visual or driven form, such as a monitor interface or printer interface. The processor 101 calls the computer program in the memory 102, and when the processor 101 executes the computer program, it implements the in-hospital mortality risk assessment method provided by the present invention.

[0100] Based on the same inventive concept, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the in-hospital mortality risk assessment method provided by the present invention.

[0101] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0102] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented directly using hardware. Based on this understanding, the above technical solutions of the present invention, in essence or in other words, the part that contributes to the prior art, can be embodied in the form of a software product; this computer software product can be stored in a computer-readable storage medium, which can be a disk, solid-state storage, server storage, cloud object storage, or other non-transitory storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and contains several instructions that can be used to cause a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or parts thereof of the present invention.

[0103] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific application examples. It should be understood that the specific application examples described herein are merely illustrative and not intended to limit the invention.

[0104] Example 1: Calculation for low-risk patients

[0105] A patient was admitted with four indicators: ALB = 30 g / L, PCT = 0.2 ng / mL, BUN = 4.0 mmol / L, and WBC = 3.0 × 10^9 / L. None of these indicators reached the risk threshold, so A = P = U = B = 0. z = -4.598, Pr = 1 / (1 + e^{4.598}) ≈ 0.010, or approximately 1.0%. The system outputs "low risk," and the 0.10 early warning is not triggered.

[0106] Example 2: Calculation of intermediate-risk / early warning patients

[0107] A patient was admitted with four indicators: ALB = 20 g / L, PCT = 1.5 ng / mL, BUN = 5.0 mmol / L, and WBC = 4.0 × 10^9 / L. Therefore, A = 1, P = 1, U = 0, and B = 0. z = -4.598 + 2.022 + 1.166 = -1.410, Pr ≈ 0.196. The system outputs "Medium Risk," and a yellow early warning is triggered because Pr ≥ 0.10, suggesting enhanced monitoring and review of etiological examination.

[0108] Example 3: Calculation of high-risk patients

[0109] A representative high-risk patient's admission parameters were: ALB = 17.9 g / L, PCT = 22.63 ng / mL, BUN = 31.41 mmol / L, and WBC = 24.56 × 10^9 / L. All four parameters met the threshold, A=P=U=B=1. z = -4.598 + 2.022 + 1.166 + 1.189 + 1.467 = 1.246, Pr ≈ 0.777. The system output "high risk," triggering a red alert, and showed that low ALB, high BUN, high WBC, and high PCT all contributed to the increased risk.

[0110] Example 4: Integration with Hospital LIS / EMR

[0111] The system uses the inpatient number or visitor number as an index to read ALB, PCT, BUN, and WBC results within the admission time window from the LIS, and admission time and patient identification from the EMR. The standardization module handles unit conversion and time sorting; when all four data points are complete, the system automatically calculates the probability and writes it back to the clinical workstation. When an alert threshold is reached, the system alerts via pop-ups, the message center, or ward list icons, while retaining the original values, model version, and calculation time. If data is missing or the unit cannot be identified, the system does not generate a final risk indicator but instead displays a message indicating that additional testing or manual confirmation is required.

[0112] Example 5: Implementation on a standalone webpage or mobile device

[0113] Users input four numerical values ​​and select units on a webpage or mobile device. The front end performs format checks, while the back end performs unit standardization and probability calculations. The output page displays the predicted probability, risk level, threshold status of the four indicators, main contribution, and disclaimer. To avoid patient privacy leaks, names can be omitted, and results can be saved only as anonymous IDs; for long-term follow-up, access control and an encrypted database are used for storage.

[0114] Example 6: External Calibration and Upgrade of the Model

[0115] When deploying in new hospitals or populations, retain the definitions and thresholds of the four indicators, calibrate the model intercept using a local cohort, and update the coefficients as necessary. Save the new model as a separate version, without overwriting the original. The system should be able to save both the development version and the locally calibrated version simultaneously, and indicate the version used in each output.

[0116] The above specific embodiments are merely explanations of the present invention and are not intended to limit the present invention. After reading this specification, those skilled in the art can make modifications to these embodiments without contributing any inventive step, but as long as they are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. A method for assessing the risk of nosocomial death of AIDS-related P. marneffei based on the A-PUB scoring model, characterized in that, Includes the following steps: S1. Obtain the test values ​​of the patient's four indicators: ALB, PCT, BUN, and WBC. S2. Standardize and convert the units of the test values ​​for the four indicators; S3. Generate corresponding variables A, P, U, and B from the standardized test values ​​according to the preset binarization threshold; S4. Calculate the linear prediction value z and the in-hospital mortality prediction probability Pr using the A-PUB scoring model. The formula for calculating the linear prediction value is z = -4.598 + 2.022 × A + 1.166 × P + 1.189 × U + 1.467 × B. The formula for calculating the in-hospital mortality prediction probability is Pr = 1 / (1 + e^{-z}), where e is a natural constant. S5. Compare the predicted probability of in-hospital mortality with the preset risk level classification thresholds to obtain the corresponding risk levels.

2. The method for assessing the risk of nosocomial death of AIDS-related P. marneffei according to claim 1, wherein, The standard units for the four indicators ALB, PCT, BUN, and WBC are g / L, ng / mL, mmol / L, and ×10^9 / L, respectively.

3. The method for assessing the risk of nosocomial death of AIDS-related P. marneffei according to claim 1, wherein, The calculation formulas for A, P, U, and B are respectively: A=I(ALB<22 g / L), P=I(PCT≥1.0 ng / mL), U=I(BUN≥6.0 mmol / L), and B=I(WBC≥5.0×10^9 / L), where I(·) is an indicator function. When the condition in parentheses is true, the function is assigned a value of 1; otherwise, it is assigned a value of 0.

4. The method for assessing the risk of in-hospital mortality from AIDS-related Marneffei infection based on the A-PUB scoring model according to claim 1, characterized in that, The risk levels include: low risk level (Pr < 0.05), medium risk level (0.05 ≤ Pr < 0.20), and high risk level (Pr ≥ 0.20).

5. The method for assessing the risk of in-hospital mortality from AIDS-related Marneffei infection based on the A-PUB scoring model according to claim 1, characterized in that, In step S5, the method further includes comparing the predicted probability of in-hospital death with a preset early warning threshold to obtain a corresponding early warning indicator. The early warning indicator includes a yellow warning with Pr≥0.10 and a red high-risk warning with Pr≥0.

20.

6. A risk assessment system for in-hospital mortality from AIDS-related Marneffei infection based on the A-PUB scoring model, characterized in that, It includes a data acquisition module, a data standardization module, a threshold coding module, a probability calculation module, and a risk stratification and early warning output module; The data acquisition module is used to obtain the test values ​​of four patient indicators: ALB, PCT, BUN, and WBC. The data standardization module is used to convert the original units of the acquired test values ​​into preset standardized units. The threshold encoding module is used to generate corresponding variables A, P, U, and B from the standardized test values ​​according to four preset thresholds based on binarization rules. The probability calculation module is used to call variables A, P, U, and B to calculate the linear prediction value z and the in-hospital mortality prediction probability Pr through the A-PUB scoring model. The formula for calculating the linear prediction value is z = -4.598 + 2.022 × A + 1.166 × P + 1.189 × U + 1.467 × B; the formula for calculating the in-hospital mortality prediction probability is Pr = 1 / (1 + e^{-z}), where e is a natural constant. The risk stratification and early warning output module is used to compare the in-hospital mortality prediction probability Pr with the preset risk stratification threshold and the preset early warning threshold, and output the corresponding risk stratification and early warning indicator.

7. The in-hospital mortality risk assessment system according to claim 6, characterized in that, The four patient indicators are obtained from LIS, EMR, or manually entered. The data standardization module also includes time window filtering, duplicate value removal, reasonable range checking, and missing value indication for the obtained test values. The standard units for the four indicators ALB, PCT, BUN, and WBC are g / L, ng / mL, mmol / L, and ×10^9 / L, respectively. The calculation formulas for A, P, U, and B are A=I (ALB < 22 g / L), P=I (PCT ≥ 1.0 ng / mL), and U=I (BUN ≥ 6.0 ng / mL), respectively. The risk levels are defined as follows: mmol / L), B = I(WBC ≥ 5.0 × 10^9 / L), where I(·) is an indicator function; the function is assigned a value of 1 when the condition in parentheses is true, and a value of 0 otherwise. The risk levels include: low risk level (Pr < 0.05), medium risk level (0.05 ≤ Pr < 0.20), and high risk level (Pr ≥ 0.20). The early warning indicators include: yellow warning (Pr ≥ 0.10) and red high-risk warning (Pr ≥ 0.20).

8. The in-hospital mortality risk assessment system according to claim 7, characterized in that, The system also includes an interpretation module, a display and push module, and a version and audit module; the interpretation module is used to output the contribution of each variable to the prediction result; the display and push module is used to display the result on the doctor's workstation, ward screen, web page, mobile terminal or report; the version and audit module is used to manage model version, threshold configuration, external calibration parameters and complete calculation log.

9. An electronic device, characterized in that, It includes at least one memory, a processor, a communication interface, a display or output interface, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the steps of the in-hospital mortality risk assessment method as described in any one of claims 1-5.

10. A computer-readable medium, characterized in that, It stores a computer program, characterized in that, when executed by a processor, the computer program implements the steps of the in-hospital mortality risk assessment method as described in any one of claims 1-5; the computer-readable storage medium is a disk, solid-state storage, server storage, cloud object storage, or a non-transitory computer-readable medium.

Images