CN115691807A

CN115691807A - Slow-acceleration acute hepatic failure infection risk early warning model and construction method thereof

Info

Publication number: CN115691807A
Application number: CN202211334544.2A
Authority: CN
Inventors: 李涛
Original assignee: Shandong Provincial Hospital Affiliated to Shandong First Medical University
Current assignee: Shandong Provincial Hospital Affiliated to Shandong First Medical University
Priority date: 2022-10-28
Filing date: 2022-10-28
Publication date: 2023-02-03

Abstract

A chronic acute hepatic failure infection risk early warning model and a construction method thereof comprise the steps of collecting clinical data, carrying out descriptive statistics on the clinical data, determining clinical indexes, applying Lasso regression to analyze the clinical indexes, screening out risk factors according to the standard that the Lasso regression coefficient is less than 0.05, using multi-factor Logistic regression to analyze the risk factors, screening out independent risk factors according to the standard that the Logistic regression coefficient p is less than 0.05, constructing a risk model based on the independent risk factors, screening out an optimal risk model from the risk model, inputting parameters into the optimal risk model, generating a line graph of the parameters, finding integrals corresponding to the parameters according to the line graph, finding risk probabilities in a probability axis after counting the integrals, carrying out visual conversion on the optimal risk model, optimizing the clinical visual application of the optimal risk model, and providing a chronic acute hepatic failure infection risk early warning signal for a patient.

Description

Slow-acceleration acute hepatic failure infection risk early warning model and construction method thereof

Technical Field

The invention relates to the technical field of prediction model construction, in particular to a slow-plus acute hepatic failure infection risk early warning model and a construction method thereof.

Background

Chronic plus Acute Liver Failure (ACLF) is an Acute decompensation of Liver function caused by various causes on the basis of Chronic Liver disease, which can cause multiple organ Failure and high short-term mortality. Infection is one of the most major causes and complications of ACLF, with an incidence rate of 50-70% and one of the causes of poor prognosis of ACLF. Accurate assessment of infection risk, early diagnosis, and effective control of infection are therefore critical to reducing ACLF mortality and improving its prognosis.

In recent years, a number of risk factors have been identified in association with the development of infection in patients with ACLF. Elevated C-reactive protein (CRP), advanced hepatic encephalopathy, and elevated White Blood Cell (WBC) counts have been reported to be independently associated with the development of infection in patients with ACLF, where elevated CRP is an accurate predictor of bacterial infection in patients with ACLF associated with autoimmune liver disease. Meanwhile, some studies establish models for predicting the development and prognosis of ACLF infection based on immune inflammation and liver function indexes, but the prediction effects of the models are limited.

Disclosure of Invention

The invention provides a chronic acute hepatic failure infection risk early warning model and a construction method thereof.

The technical scheme of the invention is as follows:

a construction method of a chronic acute hepatic failure infection risk early warning model comprises the following steps:

determining clinical indexes: collecting clinical data, performing descriptive statistics on the clinical data, and determining clinical indexes;

screening risk factors: analyzing the clinical index by using a Lasso regression, and screening out risk factors according to the standard that the Lasso regression coefficient is less than 0.05;

screening independent risk factors: analyzing the risk factors by using multi-factor Logistic regression, and screening out independent risk factors according to the standard that a Logistic regression coefficient p is less than 0.05;

screening an optimal risk model: and constructing a risk model based on the independent risk factors, and screening out an optimal risk model from the risk models.

Preferably, the operation of collecting the clinical indexes includes establishing a research queue, dividing the clinical data into a training set and an external validation set, further dividing the training set and the external validation set into an infected group and a non-infected group, and dividing the training set and the external validation set into the infected group and the non-infected group, so as to facilitate description analysis and comparison of the clinical indexes, wherein the training set can be used for practice training of the model, and the external validation set can be used for further validating the prediction accuracy and calibration performance of the model.

Preferably, the operation of screening the independent risk factors comprises performing Spearman correlation analysis on the independent risk factors, and detecting the multiple correlation of the independent risk factors, so as to detect the feasibility of constructing a risk model by the independent risk factors.

Furthermore, the operation of screening the optimal risk model comprises the steps of adopting a method of increasing and decreasing the risk factors to construct the risk model, so that the risk model formed by combining different risk factors can be conveniently analyzed, and the optimal risk can be screened.

Furthermore, the operation of screening the optimal risk model comprises the steps of applying Hosmer-Lemeshow to test the calibration performance of the risk model after the risk model is built, so that the calibration performance of a plurality of risk models in the aspect of predicting the infection risk of the ACLF patient can be detected, and the optimal risk model can be further screened conveniently.

Furthermore, the operation of screening the optimal risk model further comprises the steps of evaluating the risk model by adopting a net weight reclassification index after the calibration performance of the risk model is checked by applying the Hosmer-Lemeshow, screening the optimal risk model, conveniently comparing the prediction accuracy among the risk models, and being beneficial to screening the optimal risk model.

Preferably, after the operation of screening the optimal risk model, the method further comprises dividing an application range by taking the slow and acute liver failure stages as grouping standards, analyzing the optimal risk model in the application range by applying ROC curve hierarchical statistics, and screening the optimal application range of the optimal risk model, so that the optimal risk model can be further applied.

The invention also provides application of the chronic acute liver failure infection risk early warning model, which comprises the steps of inputting parameters into the optimal risk model, generating a lineup graph (Nomogram) of the parameters, performing visual conversion on the optimal risk model, and further optimizing the clinical visual application of the model.

Furthermore, an integral corresponding to the parameter is found according to the lineogram, after the integral is counted, the risk probability is found in a probability axis, and therefore an early warning signal is provided for the infection risk of the ACLF patient conveniently.

Further, the parameters include white blood cells, urea nitrogen, and D-dimer.

The invention has the beneficial effects that: the invention relates to a slow-plus-acute liver failure infection risk early warning model and a construction method thereof.

Drawings

The aspects and advantages of the present application will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention.

In the drawings:

FIG. 1 is a flowchart of an ACLF infection early warning model and a visual construction method thereof in an embodiment;

FIG. 2 is a probability calibration graph of a risk model in an embodiment, A-an application probability calibration graph of WLBD in a training set, B-an application probability calibration graph of WBD in a training set, C-an application probability calibration graph of WD in a training set, D-a clinical decision graph of WLBD, WBD, and WD in a training set, E-a clinical impact graph of WLBD in an training set, F-a clinical impact graph of WBD in an training set, G-a clinical impact graph of WD in a training set, and H-a net reclassification index profile of WD and WBD in an training set;

FIG. 3 is a five-fold cross-validation plot of the ROC curve for WBD in the example, A-WBD-1, B-WBD-2, C-WBD-3, D-WLBD-4;

FIG. 4 is a ROC graph of WBD under different groups, a ROC graph of WBD in the A-alcohol correlation ACLF group, a ROC graph of WBD in the B-HBV correlation ACLF group, a ROC graph of WBD in the C-autoimmune liver disease correlation ACLF group, a ROC graph of WBD in the D-ACLF early stage group, and a ROC graph of WBD in the E-ACLF late stage group in the examples;

FIG. 5 is a Nomogram of WBDs in an example.

Detailed Description

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings.

The embodiment provides a chronic acute liver failure infection risk early warning model and a construction method thereof, and referring to fig. 1, the steps are as follows:

screening independent risk factors: analyzing the risk factors by using multi-factor Logistic regression, and screening out independent risk factors according to the standard that the Logistic regression coefficient is less than 0.05;

screening an optimal risk model: constructing a plurality of risk models based on the independent risk factors, and screening out an optimal risk model;

the method specifically comprises the following steps:

s1, collecting clinical indexes

The operation of collecting infection indexes comprises establishing a research queue, dividing clinical data into a training set and an external verification set, further dividing the training set and the external verification set into an infection group and a non-infection group, then cleaning and sorting the clinical data, then performing descriptive statistics on the clinical data, and determining the clinical indexes, specifically:

establishing a research queue includes establishing a training set and an external validation set

(1) Training set: patients with confirmed ACLF were collected retrospectively, and patients were divided into infected and uninfected groups based on inclusion exclusion criteria and cohort study design (see table 1), training sets were established, and internal validation sets were established on a randomized assignment basis.

(2) External verification set: patients with confirmed ACLF were collected retrospectively, and patients were divided into infected and uninfected groups according to inclusion exclusion criteria and cohort study design, and an external validation set was established.

TABLE 1

The diagnosis standard of the ACLF refers to the ACLF consensus established by the Alta liver research institute (APASL) and the Chinese liver failure diagnosis and treatment guideline standard in 2018, and specifically comprises the following steps:

the ACLF consensus established by reference to the asia pacific liver research society (APASL) is:

(1) clinical manifestations of acute liver function decompensation and liver failure occur in a short time on the basis of chronic liver disease;

(2) jaundice rapidly deepens, and bilirubin rises to more than or equal to 5mg/dL (more than or equal to 85 mu mol/L);

(3) the international normalized ratio of prothrombin time (INR) is more than or equal to 1.5;

(4) ascites and/or hepatic encephalopathy occurred within 4 weeks.

According to 2018 Chinese diagnosis and treatment guideline standards for liver failure:

(1) extreme weakness, and obvious anorexia, abdominal distention, nausea, vomiting and other serious digestive tract symptoms;

(2) the jaundice is rapidly deepened, and the total bilirubin in the serum is more than or equal to 10 times of the upper limit of the normal value or the daily rise is more than or equal to 17.1 mu mol/L);

(3) has bleeding with prothrombin activity less than 40% (or INR greater than 1.5), and excludes other reasons.

Inclusion criteria for infection reference:

(1) SBP: the PMN cell count in the ascites is more than or equal to 250/mm ³ 。

(2) Pulmonary infection; chest x-rays show clinical signs of infection and new infiltration.

(3) Urinary infections; urinary sediment abnormalities (> 5/HP), positive in urine culture, negative in urine culture but unclear in leukocyte count.

(4) Skin soft tissue infections; clinical signs of infection associated with skin swelling, erythema, fever and tenderness

(5) Fungal infections; blood culture or other specimen culture with clear evidence of fungal infection

Clinical data cleaning and sorting

Using the medical record system, clinical information is collected for the patients in the cohort, including general demographic information: age, sex, clinical symptoms and signs, whether ascites exists, whether chronic viral hepatitis B exists, alcoholic liver disease, other chronic liver diseases; laboratory blood routine: red Blood Cells (RBCs), white Blood Cells (WBCs), platelets (platlets, PLTs), hemoglobin (HGBs), neutrophil percentage (neutrophiles%, NEU%), lymphocyte percentage (lymphocytes%, LYMPH%), monocyte percentage (Monocyte%, MON%), neutrophil count (NEU), lymphocyte count (lymphyte count, LYM), monocyte count (Monocyte count, MON), liver function index: alanine Aminotransferase (ALT), glutamic-oxaloacetic transaminase (AST), glutamyl transpeptidase (GGT), alkaline phosphatase (ALP), albumin (Albumin, ALB), total Bilirubin (TBIL), direct Bilirubin (DBIL), indirect Bilirubin (IBIL), urea nitrogen (Blood nitrate, BUN), creatinine (CRE), blood coagulation index: prothrombin Time (PT), prothrombin time activity (PT%), international normalized ratio of Prothrombin time (INR), D-dimer (D-dimer), thromboplastin time (APTT), fibrinogen (Fib).

Descriptive statistics of clinical data

Descriptive statistical analysis of clinical data was performed using SPSS statistics 26.0 and Rstudios 4.2.0 statistical software. Wherein, the classified variable data is represented by a counting method and a composition ratio, the continuous variable data is subjected to normality test, the data which accords with normal distribution is represented by mean +/-standard deviation, and the data which does not accord with normal distribution is represented by median and quartile range (IQR).

Wherein: (1) comparing normal distribution data groups by adopting a t test (Student's t test); (2) comparing among the non-normal distribution data groups by adopting Mann-Whitney U test; (3) the significance comparison of the ratio, the rate and the composition ratio adopts a chi-square test or a Fisher's exact test. Correlation between two continuous variable indexes is tested by using the Person correlation test. A regression coefficient p <0.05 is a significant statistical difference.

In this example, applicants have divided the clinical data according to inclusion and exclusion criteria, including 125 patients with ACLF in the training set, with an infected group (sample number n =53, proportion 42.4%) and an uninfected group (sample number n =72, proportion 57.6%), and 60 patients with an external validation set, with an infected group (sample number n =23, proportion 38.3%) and an uninfected group (sample number n =37, proportion 61.7%).

Performing a descriptive statistical analysis on the clinical data of the training set and the external validation set patients.

The applicant found, in the descriptive statistical analysis of clinical data, that in a training set of 125 patients of the training set (male: 99; female: 26), the statistical results show: ascites (with ascites (49 (68.1%) vs.48 (90.6%)); no ascites (23 (31.9%)) vs.5 (9.4%)) p < 0.05), WBC counts (5.60 (3.92, 8.22) vs.9.24 (6.07, 13.16) p < 0.001), LYMPH% (23.70 (15.57, 32.00) vs.14.30 (8.10, 22.40) p < 0.001), NEU% (65.55 (56.35, 72.85) vs.74.80 (65.10, 84.50) p < 0.001), MON counts (0.59 (0.40, 0.75) vs.0.82 (0.43, 1.22) p < 0.05), NEU counts (3.37 (2.30, 5.47.47) vs.6.70) p < 0.07, 6.10) p < 0.07); ALP (183.00 (153.00, 220.25) vs.153.00 (118.00, 221.00) p < 0.05), IBIL (102.20 (85.76, 126.60) vs.119.10 (92.90, 154.79) p < 0.05), D-dimer (1.95 (0.66, 3.30) vs.3.62 (1.83, 5.55) p < 0.001); BUN (3.90 (3.30, 5.93) vs.7.00 (4.70, 11.20) p < 0.001), CRE (62.15 (52.95, 76.25) vs.69.00 (57.00, 94.00) p < 0.05).

The 11 clinical indexes have obvious statistical difference between patients in an infection group and patients in a non-infection group, and feasibility is provided for screening infection risk factors of the patients with ACLF in the next step and evaluating the infection risk of the ACLF.

S2, screening risk factors

Specifically, lasso regression analysis is applied, regression coefficients are selected as significance difference standards, and risk factors are screened from clinical indexes of patients in an infected group and patients in a non-infected group. Specifically, the 11 clinical indexes in the S1 are analyzed by Lasso regression, and risk factors are screened out.

Further, a Lasso regression coefficient p <0.05 is selected as the significance difference criterion, and specifically, when the Lasso regression coefficient p of the clinical data is <0.05, the corresponding clinical index is a risk factor.

The results show that WBC counts (p = 0.019702014), LYMPH% (p = -0.012385514), BUN (p = 0.019168785), D-dimer (p = 0.008482463) have non-zero coefficients in the Lasso regression analysis, ultimately determining as a risk factor for infection in ACLF patients.

S3, screening independent risk factors

And analyzing the risk factors by using Logistic regression, screening out independent risk factors, and specifically, bringing the risk factors screened out by the Lasso regression analysis into the Logistic regression analysis.

In the screening of the independent risk factors using Logistic regression analysis, the risk factors are independent risk factors when the Logistic regression coefficient p of the risk factors is less than 0.05.

Applicants obtained WBC (([ OR ]:1.187, 95%: ci.

Table 2: analysis of risk factors for infection in patients with ACLF by Lasso regression and multifactor Logistic in training set

OR is a ratio of ratio; CI is a confidence interval; * The difference was statistically significant; perform edge saliency)

Further, according to the 3 screened independent risk factors of the infection of the ACLF patients through Logistic regression analysis, the area under the curve (AUC) calculated by a Receiver Operating Curve (ROC) is used for testing the prediction accuracy, specificity and sensitivity of the 3 independent risk factors and LYMPH% to the infection of the ACLF patients, and the applicant obtains the following results in the practical implementation process: the AUC (specificity, sensitivity, 95% CI) of the 4 risk factors WBC, LYMPH, BUN and D-dimer were 0.746 (0.694%, 0.736%, 0.656-0.835), 0.728 (78%, 58%, 0.637-0.819), 0.733 (58%, 83%, 0.7638-0.827), 0.693 (42%, 91%, 0.600-0.786), respectively (see Table 3).

TABLE 3 AUROC (95% confidence interval) specificity and sensitivity of WBC, LYMPH%, BUN and D-dimer for assessment of risk of infection in patients with ACLF.

Further, using Spearman correlation analysis to examine pairwise correlations of WBC, BUN and D-dimer, applicants obtained a positive correlation of WBC levels with BUN levels in actual practice (r =0.428,p = 0.000); WBC levels were positively correlated with D-dimer levels (r =0.100, p = 0.710); BUN is positively correlated with D-dimer levels (r =0.446, p = 0.000), which confirms their feasibility to jointly construct predictive models (r is the correlation coefficient and p is the check value).

S4, screening an optimal risk model

And constructing a plurality of risk models based on the risk factors and the independent risk factors, and screening out an optimal risk model.

Specifically, 3 simplified risk models of infection of patients with ACLF are constructed based on risk factors LYMPH% and independent risk factors WBC, BUN and D-dimer, and are named as WLBD (comprising WBC, LYMPH%, BUN and D-dimer), WBD (comprising WBC, BUN and D-dimer) and WD (comprising WBC and D-dimer).

As a preferred embodiment, using ROC curves to calculate the prediction accuracy of AUC for detecting WLBD, WBD and WD, the applicant found that in actual practice the AUC of 3 risk models of WLBD, WBD and WD was 0.802 (95% ci.

As a preferred embodiment, the advantages and disadvantages of WLBD, WBD and WD are evaluated by using a probability Calibration graph (DCA), a Clinical Decision and Clinical influence curve (Clinical influence curve), the results of the probability Calibration graph obtained by the applicant in the practical implementation process show that the evaluation results and the practical observation results of the 3 risk models have good consistency (see graphs A, B and C in FIG. 2), and the results of the Clinical Decision and Clinical influence curve show that the Clinical practicability of the 3 risk models is better (see graphs D, E, F and G in FIG. 2).

Further, the calibration performance of the WLBD, WBD and WD is tested by using the Hosmer-Lemeshow, so that the WLBD, WBD and WD have good calibration performance, and the WLBD, WBD and WD all have good calibration performance in the practical implementation process (p =0.531, p =0.7205, p =0.8121, p is a model fitting index), and compared with the WBD and WD, the WLBD has poor calibration performance in predicting the infection risk of the ACLF patient.

Combining the above results, both WBD and WD have better prediction accuracy and calibration efficiency.

Furthermore, WD is used as a traditional risk model, WBD formed by adding a new prediction factor BUN is used as a new risk model, and a Net Reclassification Index (NRI) is introduced to evaluate whether the overall prediction accuracy of the traditional model can be improved by adding the new prediction factor BUN in the traditional risk model (WD).

In actual practice, when applicants have found that the addition of BUN factors to WD results in an NRI of 0.436 (95% ci (0.021-0.824) regression coefficient p of 0.035), it can be seen that the new risk model (WBD) has a positive and significant reclassification effect on assessing infection risk in ACLF patients, i.e., WBD is the optimal risk model (see H in fig. 2).

Based on the above results, WBDs with higher prediction efficiency and fewer predictors are recommended as a new risk model for assessing infection risk in ACLF patients.

As a preferred embodiment, the prediction accuracy of WBD in the training set can be further verified by applying the five-fold cross-validation of ROC curve to verify the prediction accuracy of WBD.

Referring to fig. 3, the AUC results obtained by the applicant in the practical implementation process are shown as: WBD-1 (0.951, 95% CI; WBD-2 (0.786, 95% ci; WBD-3 (0.714, 95% ci; WBD-4 (0.753, 95% CI; WBD-5 (0.694, 95% CI; overall AUC averaged 0.780, therefore WBD showed good prediction accuracy in assessing infection risk prediction in ACLF patients.

As a preferred embodiment, the ROC curve is used to calculate AUC values in WBD for the external validation set, and the results show that: AUC was 0.885 (95% CI.

As a preferred embodiment, the WBD was tested for calibration performance in the external validation set using the Hosmer-Lemeshow, which also showed good calibration performance (p = 0.0614).

Furthermore, the optimal application range of the WBD is determined by taking the stage of the chronic liver failure and the acute liver failure as a layering standard and applying ROC curve layering statistics to analyze the WBD.

Specifically, the clinical manifestations of patients with chronic acute hepatic failure are divided into early stage, middle stage and late stage according to the following division criteria:

in the early stage: (1) Extreme weakness, and severe digestive tract symptoms such as obvious anorexia, vomiting, abdominal distension and the like; (2) Alanine Aminotransferase (ALT) and/or aspartate Aminotransferase (AST) are greatly increased, and jaundice is progressively deepened (TBil is more than or equal to 85.5 and less than 171 mu mol/L) or the daily increase is more than or equal to 17.1 mu mol/L; (3) bleeding tendency, 40% < PTA ≦ 50% (INR < 1.5).

Early stage: (1) Extreme weakness, and severe digestive tract symptoms such as obvious anorexia, vomiting, abdominal distension and the like; (2) ALT and/or AST continue to rise greatly, and jaundice progressively deepens (TBil is more than or equal to 171 mu mol/L or the daily rise is more than or equal to 17.1 mu mol/L); (3) Bleeding tendency, 30% < PTA < 40% (or 1.5 < INR < 1.9); (4) no complications and other extrahepatic organ failure.

In the middle stage: on the basis of early-stage liver failure manifestation, the disease condition further develops, ALT and/or AST rapidly decreases, TBil continuously increases, bleeding manifestation is obvious (bleeding point or ecchymosis), 20% < PTA < 30% (or 1.9 < INR < 2.6), and 1 complication and/or 1 extrahepatic organ failure are accompanied.

And (3) in the late stage: on the basis of the middle-stage expression of liver failure, the condition of the liver is further aggravated, the liver has severe bleeding tendency (ecchymosis and the like at an injection part), the PTA is less than or equal to 20 percent (or the INR is more than or equal to 2.6), and more than 2 complications and/or more than 2 extrahepatic organ failure appear.

Second, we divided into three subgroups, according to the etiology of chronic plus acute liver failure: (1) alcohol correlation ACLF; (2) HBV-related ACLF; (3) autoimmune liver disease-related ACLF. Based on the stage of the course of chronic plus acute liver failure, we divided ACLF patients into two groups: ACLF early stage group (prophase and early), ACLF late stage group (metaphase and late).

Finally, ROC curves are applied to calculate AUC values of WBDs in the above packets, respectively.

In practical implementation, the AUC areas of WBD array plots of three subgroups are obtained by the applicant as follows: 0.868 (95% ci; the AUC areas of WBD in the early stage and late stage ACLF groups were 0.873 (95% ci.

In addition, the embodiment also provides an application of the chronic acute liver failure infection risk early warning model, which includes inputting parameters into the optimal risk model, generating a Nomogram of the parameters, performing visual transformation on the optimal risk model, and further optimizing the clinical visual application of the model.

Specifically, three parameters of WBC, BUN and D-dimer of a patient are input into the WBD, and a Nomogram is obtained by applying an rms software package in Rstudio4.2.0 software, wherein the Nomogram is considered as a reliable and simple prediction tool and is helpful for clinical prediction and personalized evaluation promotion.

Further, referring to fig. 5, the integral corresponding to each parameter is found according to the Nomogram, and after the total integral is calculated, the risk probability is found in the probability axis, so that an ACLF infection risk early warning signal is provided for the patient.

The chronic acute liver failure infection risk early warning model and the construction method thereof provided by the invention are only used for infection risk prediction and are not used for clinical disease diagnosis.

The technical solutions provided by the present application are introduced in detail, and specific examples are applied in the description to explain the principles and embodiments of the present application, and the descriptions of the above examples are only used to help understanding the method and the core ideas of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, the specific implementation and the application range may be changed. In view of the above, the description should not be taken as limiting the application.

All the above-mentioned optional technical solutions can be combined arbitrarily to form the optional embodiments of the present invention, and are not described herein again.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. A construction method of a chronic acute hepatic failure infection risk early warning model is characterized by comprising the following steps:

determining the clinical index: collecting clinical data, performing descriptive statistics on the clinical data, and determining clinical indexes;

screening risk factors: analyzing the clinical index by using Lasso regression, and screening out risk factors according to the standard that the Lasso regression coefficient is less than 0.05;

screening independent risk factors: analyzing the risk factors by using multi-factor Logistic regression, and screening out independent risk factors according to the standard that the Logistic regression coefficient p is less than 0.05;

2. The method of claim 1, wherein the operation of collecting clinical indicators comprises establishing a research queue, dividing clinical data into a training set and an external validation set, and further dividing the training set and the external validation set into an infected group and a non-infected group.

3. The method for constructing a chronic acute liver failure infection risk early warning model according to claim 1, wherein the operation of screening independent risk factors comprises performing Spearman correlation analysis on the independent risk factors and detecting multiple correlations of the independent risk factors.

4. The method for constructing a chronic acute liver failure infection risk early warning model according to claim 1, wherein the operation of screening the optimal risk model comprises constructing the risk model by increasing or decreasing the risk factors.

5. The method for constructing a chronic acute liver failure infection risk early warning model according to claim 4, wherein the operation of screening the optimal risk model comprises applying Hosmer-Lemeshow to check the calibration performance of the risk model after constructing the risk model.

6. The method for constructing a chronic acute liver failure early warning infection risk model according to claim 5, wherein the operation of screening the optimal risk model further comprises screening the optimal risk model by evaluating the risk model with a net reclassification index after checking the calibration of the risk model by applying Hosmer-Lemeshow.

7. The method for constructing the slow-plus-acute hepatic failure early warning infection risk model according to claim 1, wherein the operation of screening the optimal risk model further comprises dividing an application range by using the slow-plus-acute hepatic failure stage as a grouping standard, analyzing the optimal risk model in the application range by applying ROC curve hierarchical statistics, and screening the optimal application range of the optimal risk model.

8. The application of the slow-plus-acute liver failure infection risk early warning model is characterized by comprising the steps of inputting parameters into the optimal risk model and generating a lineup graph of the parameters.

9. The application of the early warning model for the risk of chronic plus acute liver failure infection as claimed in claim 8, comprising finding the integral corresponding to the parameter according to the line graph, and finding the risk probability in the probability axis after counting the integral.

10. The use of the chronic acute liver failure infection risk pre-warning model according to claim 9, wherein the parameters comprise leukocytes, urea nitrogen and D-dimer.