CN118888152A - A LATI prognostic model for patients with liver failure and its construction method - Google Patents

Abstract

The invention discloses LATI prognosis model for liver failure patients and a construction method, wherein the model comprises four evaluation indexes: age, total bilirubin, INR and lactic acid, said indices forming a scoring system for predicting 28-day risk of death and 90-day risk of death for a liver failure patient, wherein predicting 28-day risk of death for a liver failure patient comprises the calculation formula: 28-day score = -7.04+0.21-lactic acid+0.04-age+0.01-total bilirubin+0.69-inr, predicting 90-day mortality risk in liver failure patients, wherein the scoring system comprises the following calculation formula: 90-day score = -7.12+0.16+lactic acid+0.05+age+0.01+total bilirubin+0.76 inr. The liver failure prognosis model has few inclusion indexes, is suitable for all types of liver failure patients, has higher prediction capacity than the existing model, and has obvious clinical guiding significance.

Description

A LATI prognostic model for patients with liver failure and its construction method

Technical Field

The present invention belongs to the field of medical technology, and specifically relates to a LATI prognosis model for patients with liver failure and a construction method thereof.

Background Art

Liver failure is a serious liver damage caused by multiple factors, which leads to serious disorders or decompensation of synthesis, detoxification, metabolism and biotransformation functions, and a group of clinical syndromes with jaundice, coagulopathy, hepatorenal syndrome, hepatic encephalopathy, ascites and other symptoms as the main manifestations. The mortality rate of patients with liver failure is extremely high, and prognosis assessment should be carried out throughout the entire process of diagnosis and treatment, especially emphasizing the importance of early prognosis assessment. Child-Turcotte-Pugh (CTP) score and Model for end-stage liver disease (MELD) are commonly used methods to assess liver function and judge prognosis, but there are shortcomings such as some indicators are highly subjective and easily affected by non-liver disease factors.

The 2019 consensus recommendation of the Asian Pacific Association for the Study of Liver (APASL) on acute-on-chronic liver failure (ACLF) proposed that ACLF refers to acute liver injury on the basis of chronic liver disease/cirrhosis (previously diagnosed/undiagnosed), with jaundice (serum bilirubin ≥5 mg/dl) and coagulopathy [international normalized ratio (INR) ≥1.5 or prothrombin activity (PTA) <40%] as the main manifestations, and ascites and/or HE within 4 weeks. Among patients who survive 90 days, about 70% of ACLF patients gradually recover.

The APASL-ACLF Research Consortium (AARC) score includes five indicators: total bilirubin, creatinine, serum lactate, INR, and hepatic encephalopathy. The AARC score is a good tool for evaluating prognosis and can assess which patients may experience reversal. It is better than the MELD, MELD-Na, CLIF-SOFA, and SOFA scores. The cumulative mortality rate increases with the increase of the AARC score. The trend of the AARC score within the first week can predict whether liver transplantation is needed.

The Chronic Liver Failure Consortium Organ Failure (CLIF-C OF) developed the Chronic Liver Failure Consortium Acute-on-Chronic Liver Failure (CLIF-C ACLF) scoring system, which is a complex scale based on 6 organ failures and 11 predictive factors, including: liver: total bilirubin; kidney: creatinine and renal replacement therapy; neurological: hepatic encephalopathy grade; coagulation function: INR; circulation: mean arterial pressure and use of vasoactive drugs; respiratory: PaO ₂ , SpO ₂ , FiO ₂ , and use of mechanical ventilation. The MELD scoring system includes 5 indicators: serum bilirubin, creatinine (Scr), INR, and liver etiology or serum sodium. It is now widely used in the prognosis assessment of various liver diseases. The higher the score, the worse the prognosis and the higher the mortality rate. However, since serum Scr determination is affected by non-liver disease factors, the MELD score may misjudge the severity of liver disease. Since then, there have been continuous studies to improve MELD, deriving the MELD-Na, iMELD, and MELD to SNa ratio (MESO) scoring systems. In 2015, Roayaie et al. proposed the platelet-albumin-bilirubin (PALBI) scoring system, which was originally used to determine the prognosis of liver cancer. However, in recent years, many studies have shown that the PALBI score has predictive power for the prognosis of cirrhosis, and is even better than MELD. The Tongji prognostic predictor model (TPPM) score includes total bilirubin, INR, complications, and HBVDNA quantification, and is a scoring system developed for HBV-related acute-on-chronic liver failure.

None of the above existing prediction models can predict the prognosis of all liver failure. The influencing factors mainly include total bilirubin, INR, creatinine, hepatic encephalopathy and other indicators, which have strong limitations. Among them, the judgment and severity classification of hepatic encephalopathy (HE) are highly subjective. HE is a continuous manifestation from normal cognitive function and complete consciousness to coma, and mild HE is a very hidden stage in the onset of HE. Creatinine is easily affected by non-liver disease factors. It is necessary to develop a prediction model with strong objectivity and fewer interference factors (indicators) to facilitate the clinical judgment of liver failure prognosis. Therefore, the development of a prognosis assessment model that is easy to use and applicable to all patients with liver failure is an urgent problem to be solved.

Summary of the invention

Based on the defects and limitations of the existing liver failure prognosis scoring model, the present invention provides a LATI prognosis model for liver failure patients and a construction method thereof. The prognosis model for liver failure patients constructed by the present invention has influencing factor indicators, small interference factors, but strong predictive prognosis ability, and is suitable for the prognosis evaluation of all types of liver failure patients.

In one embodiment, the present invention provides a LATI prognostic model for patients with liver failure, which includes four influencing factor indicators: age, total bilirubin, INR and lactate. The influencing factors are formed into a scoring system for predicting the 28-day mortality risk and 90-day mortality risk of patients with liver failure.

Furthermore, the LATI prognostic model of the present invention can be applied to the prognosis assessment of patients with all types of liver failure.

In some embodiments, the above-mentioned LATI prognostic model, the calculation formula of the scoring system for predicting the 28-day mortality risk of patients with liver failure is: 28-day score = -7.04 + 0.21*lactate + 0.04*age + 0.01*total bilirubin + 0.69*INR; the calculation formula of the scoring system for predicting the 90-day mortality risk of patients with liver failure is: 90-day score = -7.12 + 0.16*lactate + 0.05*age + 0.01*total bilirubin + 0.76*INR.

In another embodiment, the present invention also provides a method for constructing the LATI prognostic model for liver failure patients of the present invention, comprising the following steps:

1) Recruit a sufficient sample size of patients with liver failure to serve as a development cohort;

2) Follow up the patients in the development cohort for 28 or 90 days and collect the patient's index data, including age, general information, biochemical indicators, imaging examinations and complications;

3) The indicator data of the death group and the survival group at 28 days or 90 days in the development cohort group were analyzed by univariate and multivariate logistic regression to screen out the prognostic factors with significance or influence;

4) Based on the R "caret" software package, the indicator data of one-fifth of the patients in the development cohort were randomly selected as the objects of the training set of the machine learning algorithm, and the patient indicator data were used to construct the training set. The remaining people in the development cohort were assigned to the internal validation set, and the predictive performance of the factors affecting the training results was evaluated. The Logstic regression analysis algorithm was selected for further model construction;

5) Based on logistic regression analysis, age, total bilirubin, lactate, and INR were determined to be independent influencing factors for the prognosis of patients with liver failure and used to construct a prediction model;

6) Construct the 28d prediction model formula: 28-day LATI score = -7.04 + 0.21*lactate + 0.04*age + 0.01*total bilirubin + 0.69*INR, and construct the 90-day prediction model formula: 30-day LATI score = -7.12 + 0.16*lactate + 0.05*age + 0.01*total bilirubin + 0.76*INR.

Furthermore, in the construction method of the present invention, the liver failure patients in step 1) include: HBV-related liver failure patients, acute liver failure patients and chronic liver failure patients.

Preferably, in the construction method of the present invention, the sample size of patients with liver failure described in step 1) is 1810 cases; the Logistic regression analysis score described in step 4) further includes filling the randomly missing data with a multiple interpolation method if the missing data exceeds 10% but does not exceed 50%.

Preferably, in the construction method of the present invention, the scoring in step 6) is to score the 28-day and 90-day LATI of patients with liver failure into a high-risk group and a low-risk group according to the cutoff value of the Kaplan-Meier survival curve shown in FIG6 .

In another embodiment, the prognostic model for liver failure patients of the present invention is used to predict the prognosis evaluation of all types of liver failure patients at 28 days or 90 days. Preferably, all types of liver failure patients include HBV-related liver failure patients, acute liver failure patients, chronic liver failure patients or chronic acute liver failure patients.

Terminology: The LATI prediction model refers to a prediction model composed of four evaluation indicators: lactate, age, total bilirubin, and INR. LATI is named after the first letters of the English words lactate, Age, total bilirubin, and INR, respectively.

The technical effects of the prognostic model for patients with liver failure constructed by the present invention are as follows:

1. In all cohorts, the LATI score had a better predictive effect than the existing liver failure models (AARC, CTP, MELD, PALBI, MELD-Na, TPPM, ABIC, iMELD, MESO, CLIF-C OF, CLIF-C ACLF scores), showing good discrimination and calibration.

2. The LATI score in the development cohort and internal validation cohort was superior to the prediction models for 28-day and 90-day mortality in acute-on-chronic liver failure, acute liver failure, chronic liver failure, and HBV-related liver failure, respectively.

3. The LATI score based on four indicators is an objective and reliable prediction model for all patients with liver failure. It can accurately and timely determine the 28-day and 90-day prognosis of patients with liver failure and guide clinical treatment.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of the research design of the prognostic model of the present invention.

Figure 2 is a comparison of the average AUC values of various machine learning algorithms for predicting 28-day and 90-day mortality in patients with liver failure, where A: 28-day box plot; B: 90-day box plot; C: 28-day ROC curve; D: 90-day ROC curve.

FIG3 is a nomogram for predicting 28-day and 90-day mortality in patients with liver failure in the development cohort, wherein A: 28-day mortality nomogram; B: 90-day mortality nomogram.

Figure 4 shows the comparison of the area under the ROC curve (AUC) of the LATI score prognostic model and other prognostic models for predicting the 28-day and 90-day mortality rates of patients with liver failure, where A: 28-day ROC curve of the development cohort; B: 90-day ROC curve of the development cohort; C: 28-day ROC curve of the internal validation cohort; D: 90-day ROC curve of the internal validation cohort; E: 28-day ROC curve of the external validation cohort; F: 90-day ROC curve of the external validation cohort.

Figure 5 shows the calibration ability of the LATI score prognostic model in predicting the 28-day and 90-day mortality rates of patients with liver failure, where A: 28-day calibration curve of the development cohort; B: 90-day calibration curve of the development cohort; C: 28-day calibration curve of the internal validation cohort; D: 90-day calibration curve of the internal validation cohort; E: 28-day calibration curve of the external validation cohort; F: 90-day calibration curve of the external validation cohort.

Figure 6 shows the 28-day and 90-day Kaplan-Meier survival curves based on the cutoff values of the LATI score for patients with liver failure, including: A: 28-day survival curve of the development cohort; B: 90-day survival curve of the development cohort; C: 28-day survival curve of the internal validation cohort; D: 90-day survival curve of the internal validation cohort; E: 28-day survival curve of the external validation cohort; F: 90-day survival curve of the external validation cohort.

Figure 7 is a comparison of the area under the ROC curve (AUC) of the LATI score prognostic model and other prognostic models for predicting the 28-day and 90-day mortality rates in patients with acute-on-chronic liver failure, where A: 28-day ROC line of the development cohort; B: 90-day ROC curve of the development cohort; C: 28-day ROC curve of the internal validation cohort; D: 90-day ROC curve of the internal validation cohort.

Figure 8 is a comparison of the area under the ROC curve (AUC) of the LATI score prognostic model and other prognostic models for predicting the 28-day and 90-day mortality rates of patients with acute liver failure, where A: 28-day ROC curve of the development cohort; B: 90-day ROC curve of the development cohort; C: 28-day ROC curve of the internal validation cohort; D: 90-day ROC curve of the internal validation cohort.

Figure 9 is a comparison of the area under the ROC curve (AUC) of the LATI score prognostic model and other prognostic models for predicting the 28-day and 90-day mortality rates of patients with chronic liver failure, where A: 28-day ROC curve of the development cohort; B: 90-day ROC curve of the development cohort; C: 28-day ROC curve of the internal validation cohort; D: 90-day ROC curve of the internal validation cohort.

Figure 10 is a comparison of the area under the ROC curve (AUC) of the LATI score prognostic model and other prognostic models for predicting the 28-day and 90-day mortality rates of patients with HBV-related liver failure, where A: 28-day ROC curve of the development cohort; B: 90-day ROC curve of the development cohort; C: 28-day ROC curve of the internal validation cohort; D: 90-day ROC curve of the internal validation cohort.

DETAILED DESCRIPTION

The following examples are representative and are used to further understand the essence of the present invention, but are not intended to limit the scope of the present invention in any way.

Example 1 Construction of a prognostic model for patients with liver failure

1. Patient Collection

The hospitalization data of patients diagnosed with liver failure were collected. Liver failure patients included acute liver failure, acute-on-chronic liver failure, or chronic liver failure. The inclusion criteria were as follows: the test indicators during hospitalization met the following conditions: jaundice (serum bilirubin ≥ 5 mg/dL) and coagulopathy [INR ≥ 1.5 or PTA < 40%]; the following patients were excluded: (1) patients with primary liver cancer or other malignant tumors; (2) patients with severe heart, brain, kidney, respiratory, or blood system diseases; (3) patients with portal vein thrombosis or deep vein thrombosis; (4) patients after TIPS surgery, splenectomy, or liver transplantation; (5) patients who used anticoagulant or antiplatelet drugs within the past month and had DIC; (6) patients who were pregnant; and (7) patients with HIV infection. Clinical data including general information, biochemical indexes, imaging examinations and complications of patients were collected within 24 hours after the diagnosis of liver failure in our hospital, and AARC, CTP, MELD, PALBI, MELD-Na, TPPM, ABIC, iMELD, MESO, CLIF-C OF, and CLIF-C ACLF scores were calculated. The prognostic information of patients was followed up for 28 days and 90 days, and they were divided into death group and survival group according to the prognosis. The death group included patients who died during the follow-up period, transferred to liver transplantation, and were discharged automatically due to ineffective medical treatment. All included patients received standardized comprehensive medical treatment. This study was approved by the Ethics Committee of the Second Affiliated Hospital of Chongqing Medical University, approval number: 140/2023.

2. Study Design

We collected clinical data of 6018 patients with liver failure from the Second Affiliated Hospital of Chongqing Medical University, the First Affiliated Hospital of Chongqing Medical University, and Chongqing Emergency Medical Center from 2008 to 2023, and finally included 3226 cases according to the exclusion criteria. 1810 patients from the Second Affiliated Hospital of Chongqing Medical University from 2017 to 2023 were used for model development and named the development cohort group; 1267 patients from 2008 to 2017 were mainly used for internal validation and named the internal validation cohort group; 149 patients from the First Affiliated Hospital of Chongqing Medical University and Chongqing Emergency Medical Center were used for external validation and named the external validation cohort group.

First, the predictive factors associated with 28-day and 90-day mortality were determined from the clinical data of the development cohort patients, including general information, biochemical indicators, imaging examinations, and complications, and a new prognostic score was developed. Secondly, the other two data sets were used for internal and external validation to evaluate the predictive effect of the new scoring system. Finally, patients with acute-on-chronic liver failure, acute liver failure, chronic liver failure, and HBV-related liver failure in the development cohort and internal validation cohort were screened to further verify the predictive effect of the new model. The specific research design flow chart is shown in (Figure 1).

3. Machine Learning Algorithms

Based on the R "caret" software package, one fifth of the patients in the development cohort were randomly selected as the subjects of the training set, and the rest were assigned to the internal validation set. Typical machine learning algorithms include linear support vector machine (LSVM), radial basis kernel support vector machine (RBF SVM), adaptive boosting (AdaBOOST), logistic regression, artificial neural network (ANN), K-nearest neighbor (KNN) and other algorithms. The training set was constructed with the selected variables using the default settings. These models were trained 50 times, and their predictive performance was evaluated using the internal validation set. The receiver operating characteristic (ROC) curve and the area under the curve (AUC) were used to measure the discrimination of the models constructed by different algorithms. The comparison of AUC was performed using the DeLong test. The algorithm that showed the best performance in the internal validation group was selected and used for further model construction.

4. Statistical methods

For randomly missing data, if the missing data exceeded 10% but did not exceed 50%, multiple imputation was used to fill the missing data. Normally distributed measurement data were expressed as mean ± standard deviation, and the comparison between the two groups was performed by t test. Non-normally distributed measurement data were expressed as median (interquartile range), and the comparison between the two groups was performed by Wilcoxon Mann-Whitney test, and the comparison between multiple groups was performed by Kruskal-Wallis rank sum test; the chi-square test was used for intergroup comparison of count data. Spearman correlation analysis was used to test the correlation between lactate and infection, clinical complications, and immune indicators. Univariate and multivariate logistic analysis was used to analyze the influencing factors of the prognosis of patients at 28 days and 90 days in the development cohort, and a new model LATI score for the prognosis prediction of patients with liver failure was constructed, and a nomogram for the risk of death was drawn. The receiver operating characteristic (ROC) curve and area under the curve (AUC) were used to measure the discrimination of the new model in the development and validation cohorts, and the DeLong test was used to compare the AUC. The calibration curve was used to measure the calibration of the new model. Kaplan-Meier plots were drawn according to the cutoff values of the new model, and the differences in survival rates among the groups were compared using Breslow test and univariate COX regression.

5. Baseline characteristics and model algorithm selection

A total of 3226 patients with liver failure who met the criteria were screened in three centers between 2008 and 2023. Table 1 summarizes the characteristics of patients in the development cohort and validation cohort. The mortality rate of the internal validation cohort (28d 42.9%, 90d 48.3%) was significantly higher than that of the development cohort (28d 28.5%, 90d 35.5%) and the external validation cohort (28d 22.8%, 90d 37.6%). The median age of patients in the development and internal validation cohorts was 49 years (interquartile range, 40-56), and the average age of patients in the external validation cohort was 47.6 years (standard deviation, 12.5), and male patients were the majority. The most common complications were spontaneous bacterial peritonitis and ascites. Since the lactate values were collected from arterial blood gas analysis in the First Affiliated Hospital of Chongqing Medical University, the median lactate value in the external validation cohort was smaller than that in the development and internal validation cohorts.

Among all the models, logistic regression showed the best average performance in the internal validation group and was used for further optimization. The AUC area of the six algorithms for predicting 28-day mortality in patients with liver failure was ranked as follows: logistic regression > LSVM > ANN > AdaBOOST > RBF SVM > KNN. The differences between logistic regression and the other five machine learning algorithms were statistically significant (Delong test P value < 0.05) (Figure 2A, C). The AUC area for predicting 90-day mortality in patients with liver failure was ranked as follows: logistic regression > LSVM > ANN > AdaBOOST > RBF SVM > KNN. The differences between logistic regression and the other five machine learning algorithms were statistically significant (Delong test P value < 0.05) (Figure 2B, D).

Table 1. Baseline characteristics of patients with liver failure at admission in the development and validation cohorts

*Mean ± standard deviation. Median (interquartile range). t-test. Wilcoxon Mann-Whitney test.§c ² test. Kruskal-Wallis rank sum test.

6. Construction of LATI scoring model

The results of logistic regression analysis showed (Table 2): age, hepatic encephalopathy, ascites, gastrointestinal bleeding, septic shock, total bilirubin, albumin, lactate, INR, and urea nitrogen were independent influencing factors for the 28-day prognosis of patients with liver failure (all P values were <0.05); age, hepatic encephalopathy, ascites, gastrointestinal bleeding, septic shock, total bilirubin, albumin, and INR were independent influencing factors for the 90-day prognosis of patients with liver failure (all P values were <0.05). Many studies have shown that lactate is of great significance in the prognosis of patients with liver failure. Spearman correlation analysis showed that lactate was positively correlated with WBC, N, N%, CRP, PCT, and HE grade (r values were 0.40, 0.42, 0.34, 0.11, 0.23, and 0.21, respectively, and P values were all < 0.001; lactate was negatively correlated with L% (r value was -0.31, P value < 0.001) (Table 3). Considering that albumin and urea nitrogen are easily affected by non-liver disease factors and the severity of hepatic encephalopathy is highly subjective, we finally included the following 4 indicators into the new model (LATI scoring model): age, total bilirubin, INR, Lactate, the formula is as follows: 28-day Lactate-Age-TBiL-INR (LATI) scoring system = -7.04 + 0.21 * lactate + 0.04 * age + 0.01 * total bilirubin + 0.69 * INR (Figure 3A); 90-day Lactate-Age-TBiL-INR (LATI) scoring system = -7.12 + 0.16 * lactate + 0.05 * age + 0.01 * total bilirubin + 0.76 * INR (Figure 3B). The nomogram of the LATI score was used to predict the risk of death at 28 and 90 days in patients with liver failure in the development cohort.

Table 2. Univariate and multivariate logistic regression analysis of 28-day and 90-day mortality in patients with liver failure in the development cohort

Table 3. Correlation analysis between lactic acid and infection, clinical complications and immune indicators

Example 2 Predictive ability of the LATI prognostic model score of the present invention

1. Predictive ability of LATI score in the development cohort

The AUC of LATI score for predicting 28-day mortality in patients with liver failure was 0.793 (95% confidence interval [CI] 0.77-0.82), with a sensitivity of 73.2% and a specificity of 72.4%. The AUC area was ranked as follows: LATI score > CLIF-C ACLF > ABIC > CLIF-C OF > AARC > iMELD > MESO > MELD > MELD-Na > PALBI. Except for CLIF-C ACLF, the differences between LATI score and other scores were statistically significant (DeLong test P value < 0.05). The AUC of the LATI score for predicting 90-day mortality in patients with liver failure was 0.803 (95% confidence interval [CI] 0.78-0.83), with a sensitivity of 72.1% and a specificity of 75.9%. The AUC area was ranked as follows: LATI score > CLIF-C ACLF > ABIC > CLIF-C OF > iMELD > AARC > MESO > MELD-Na > MELD > PALBI. Except for CLIF-C ACLF, the differences between the LATI score and each score were statistically significant (DeLong test P value < 0.05) (Table 4, Figure 4A and B). These results show that the LATI score can well predict the 28-day and 90-day mortality of patients with all types of liver failure in the development cohort.

Table 4. Predictive ability of LATI score in the development cohort

*DeLong test

2. Predictive ability of LATI score in the internal validation cohort

Although the development and internal validation cohorts were from the same center, the conditions of the patients were different (Table 1). In the internal validation cohort, the median total bilirubin (238.4 versus 227 umol/L, P = 0.001), median INR (2.4 versus 2.1, P < 0.001), and median lactate (3.2 versus 3.0 mmol/L, P < 0.001) were higher, but the median age was younger (47 versus 50, P < 0.001). However, the performance of the LATI score was equally outstanding. The AUC of the LATI score for predicting the 28-day mortality of patients with liver failure was 0.742 (95% confidence interval [CI] 0.71-0.77), with a sensitivity of 57.2% and a specificity of 78.2%. The AUC area was ranked as follows: MESO>LATI score>iMELD>MELD>MELD-Na>ABIC>AARC>PALBI. The difference between the LATI score and the PALBI score was statistically significant (DeLong test P value <0.001). The AUC of LATI score for predicting 90-day mortality in patients with liver failure was 0.764 (95% confidence interval [CI] 0.74-0.79), with a sensitivity of 71.9% and a specificity of 68.5%. The AUC area was ranked as follows: LATI score > iMELD > ABIC > MESO > MELD = MELD-Na > AARC > PALBI. The difference between LATI score and MELD, MELD-Na, AARC, and PALBI scores was statistically significant (DeLong test P value < 0.05) (Table 5, Figure 4C and D). Although the patients in the internal validation cohort were more seriously ill, it also showed satisfactory prediction effects in terms of 28-day and 90-day mortality in patients with all types of liver failure.

Table 5. Predictive ability of LATI score in the internal validation cohort

*DeLong test

3. Predictive ability of LATI score in the external validation cohort

In the two centers participating in the external validation, 149 patients met the study criteria and had complete data (Table 1). The AUC of the LATI score for predicting 28-day mortality in patients with liver failure was 0.765 (95% confidence interval [CI] 0.67-0.86), with a sensitivity of 79.4% and a specificity of 68.7%. The AUC area was ranked as follows: LATI score>CLIF-C OF>CLIF-C ACLF>ABIC>iMELD>MESO>MELD>AARC>MELD-Na>PALBI. The differences between the LATI score and MELD-Na, iMELD, AARC, and PALBI scores were statistically significant (DeLong test P value <0.05). The AUC of LATI score for predicting 90-day mortality in patients with liver failure was 0.794 (95% confidence interval [CI] 0.72-0.8), with a sensitivity of 80.4% and a specificity of 68.8%. The AUC area was ranked as follows: LATI score > CLIF-C ACLF > ABIC > iMELD > CLIF-C OF > MESO > MELD-Na > MELD > AARC > PALBI. The difference between LATI score and AARC and PALBI scores was statistically significant (DeLong test P value < 0.05) (Table 6, Figure 4E and F). These results show that in an independent external cohort, LATI score is still the best predictor of 28-day and 90-day mortality in patients with liver failure.

Table 6. Predictive ability of LATI score in the external validation cohort

*DeLong test

4. Calibration

The calibration ability of the LATI score to predict 28-day and 90-day mortality in patients with liver failure was evaluated in the development and validation cohorts. The observed and predicted probabilities of mortality with the new score were similar in each cohort. In the development cohort, the 28-day mortality rate in patients with liver failure was 28.5%, R2=0.338, Brier=0.147; the 90-day mortality rate was 35.5%, R2=0.365, Brier=0.161 (Figure 5A and B). In the internal validation cohort, the 28-day mortality rate in patients with liver failure was 42.9%, R2=0.269, Brier=0.194; the 90-day mortality rate was 48.3%, R2=0.310, Brier=0.191 (Figure 5C and D). In the external validation cohort, the 28-day mortality rate of patients with liver failure was 22.8%, R2=0.202, Brier=0.150; the 90-day mortality rate was 37.6%, R2=0.341, Brier=0.171 (Figure 5E and F). These results show that the new scoring system has good predictive accuracy.

5. Risk Stratification

In the development and validation cohorts, the LATI score was risk stratified according to the optimal cutoff value of the ROC curve. Patients with liver failure were divided into high-risk and low-risk groups according to the 28-day and 90-day prognostic scores. In the development cohort, the survival rate of the high-risk group was significantly lower than that of the low-risk group (28 days: HR 9.75, 95% CI 6.82-13.93, P < 0.001; 90 days: HR 7.31, 95% CI 5.66-9.46, P < 0.001, Figure 6A and B). Similarly, in the internal validation cohort, the survival rate of the high-risk group was lower than that of the low-risk group (28 days: HR 4.88, 95% CI 3.57-6.68, P < 0.001; 90 days: HR 0.18, 95% CI 0.13-0.24, P < 0.001, Figure 6C and D). In the external validation cohort, the survival rate of the high-risk group was significantly lower than that of the low-risk group (28 days: HR 12.65, 95% CI 2.91-55.06, P < 0.001; 90 days: HR 8.74, 95% CI 3.38-22.58, P < 0.001, Figure 6E and F). These results show that the LATI score is a simple and effective tool to evaluate the survival differences of patients with liver failure.

Example 3 Application value of LATI score in patients with liver failure of different types or causes

AARC, CLIF-C OF, and CLIF-C ACLF are scoring systems developed for acute-on-chronic liver failure. To avoid exaggerating the shortcomings of AARC, CLIF-COF, and CLIF-C ACLF scores, we screened out patients with acute-on-chronic liver failure for further verification. According to the 2023 American College of Gastroenterology (ACG) guidelines, the MELD score is recommended as a prognostic evaluation model for ALF patients, and the MELD scoring system is a classic model for prognostic evaluation of various liver diseases. We screened out patients with acute liver failure, chronic liver failure, and HBV-related liver failure for further verification.

1. Acute-on-chronic liver failure

In the development cohort, the LATI score had the highest AUC for predicting 28-day and 90-day mortality in patients with acute-on-chronic liver failure. The AUC difference between the LATI score and the AARC and CLIF-C OF scores in predicting 28-day mortality was statistically significant (DeLong test P value < 0.05). The AUC difference between the LATI score and the AARC score in predicting 90-day mortality was statistically significant (DeLong test P value < 0.05) (Table 7, Figure 7A, B). In the internal validation cohort, the AUC area of the LATI score in predicting 28-day and 90-day mortality in patients with acute-on-chronic liver failure was greater than that of the AARC score. The AUC difference between the LATI score and the AARC score in predicting 90-day mortality was statistically significant (DeLong test P value < 0.05) (Table 7, Figure 7C, D).

Table 7. Predictive ability of LAT I score in patients with acute-on-chronic liver failure in the development and internal validation cohorts

*DeLong test

2. Acute liver failure

In the development cohort, the AUC area of the LATI score in predicting the 28-day and 90-day mortality in patients with acute liver failure was greater than that of the MELD score. The difference in AUC between the LATI score and the MELD score in predicting the 28-day mortality was statistically significant (DeLong test P value < 0.05) (Table 8, Figure 8A, B). In the internal validation cohort, the AUC area of the LATI score in predicting the 28-day and 90-day mortality in patients with acute liver failure was greater than that of the MELD score (Table 8, Figure 8C, D).

Table 8. Predictive ability of the LATI score in patients with acute liver failure in the development and internal validation cohorts

*DeLong test

3. Chronic liver failure

In the development cohort, the AUC of the LATI score for predicting 28-day and 90-day mortality in patients with chronic liver failure was greater than that of the MELD score (Table 9, Figure 9A, B). In the internal validation cohort, the AUC of the LATI score for predicting 28-day and 90-day mortality in patients with chronic liver failure was less than that of the MELD score. There was no statistically significant difference in the AUC between the LATI score and the MELD score for predicting 28-day and 90-day mortality (DeLong test P value > 0.05) (Table 9, Figure 9C, D).

Table 9. Predictive ability of LAT I score in patients with chronic liver failure in the development and internal validation cohorts

*DeLong test

4. HBV-related liver failure

In the development cohort, the LATI score had the highest AUC for predicting 28-day and 90-day mortality in patients with HBV-related liver failure. The AUCs for predicting 28-day and 90-day mortality between the LATI score and the TPPM and MELD scores were statistically significant (DeLong test P value < 0.05) (Table 10, Figure 10A, B). In the internal validation cohort, the LATI score had the highest AUC for predicting 28-day mortality in patients with HBV-related liver failure; the AUC for predicting 90-day mortality in patients with HBV-related liver failure was only less than that of the TPPM score, and the difference between the two was not statistically significant (DeLong test P value > 0.05) (Table 10, Figure 10C, D).

Table 10. Predictive ability of LATI score in patients with HBV-related liver failure in the development and internal validation cohorts

*DeLong test

These results indicate that the LATI score in the development cohort and the internal validation cohort was superior to the prediction models for predicting 28-day and 90-day mortality in acute-on-chronic liver failure, acute liver failure, chronic liver failure, and HBV-related liver failure, respectively.

Example 4 Comparison of LATI model and COSSH-ACLF II scoring model

In the development cohort, the AUC of the LATI score for predicting 28-day and 90-day mortality in patients with liver failure was lower than that of the COSSH-ACLFII score, and the difference between the two was not statistically significant (DeLong test P value＞0.05). In the internal validation cohort, the AUC of the LATI score for predicting 28-day mortality was lower than that of the COSSH-ACLF II score, and the difference between the two was not statistically significant (DeLong test P value＞0.05). In the external validation cohort, the LATI score had the highest AUC for predicting 28-day mortality in patients with liver failure; the AUC for predicting 90-day mortality in patients with liver failure was lower than that of the COSSH-ACLF II score, and the difference between the two was not statistically significant (DeLong test P value＞0.05) (Table 11).

Table 11. Comparison of the predictive ability of LATI score and COSSH-ACLF II score in patients with liver failure in all cohorts

*DeLong test

discuss

Liver failure patients are seriously ill and progress rapidly. Prognosis assessment should be carried out throughout the entire process of diagnosis and treatment, especially the importance of early prognosis assessment, because it can guide treatment and reduce mortality. Any prognosis prediction model should use objective and easily available clinical indicators to simply and accurately predict the prognosis of patients with liver failure. The present invention determines the clinical characteristics of patients with liver failure based on multi-center, large-sample clinical data and laboratory indicators, and develops a new simple scoring model, namely the LATI scoring model, which can accurately predict the 28-day and 90-day mortality rates of such patients.

According to the 2019 APASL consensus on acute-on-chronic liver failure, the AARC score is a good tool for evaluating the prognosis of acute-on-chronic liver failure. The score includes five indicators: total bilirubin, hepatic encephalopathy grade, lactate, INR, and creatinine. It can evaluate which patients may experience reversal, and the effect is better than MELD, MELD-Na, CLIF-SOFA, and SOFA scores. The cumulative mortality rate increases with the increase of the AARC score. The trend of the AARC score within the first week can predict whether liver transplantation is required: <10 points or reduced to <10 points, the survival rate is improved, and liver transplantation is not required; patients with an AARC score of >10 points should be included in the list of liver transplant candidates. AARC-ACLF should be evaluated on the 4th and 7th days to predict the progression and prognosis of the disease. Patient grading according to the AARC score [Grade I (5-7 points), Grade II (8-10 points), Grade III (11-15 points] can effectively predict and guide treatment. In general, the AARC-ACLF score is easy to use, dynamic and reliable, and is superior to existing prediction models. It can reliably predict the need for intervention within the first week, such as liver transplantation. The CLIF-C OF and CLIF-CACLF scores are also scoring systems developed for acute-on-chronic liver failure. They are complex scales based on 6 organ failures and 11 predictive factors, including: liver: total bilirubin; kidney: creatinine and renal replacement therapy; neurology: hepatic encephalopathy grade; coagulation function: INR; circulation: mean arterial pressure and use of vasoactive drugs; respiration: PaO ₂ , SpO ₂ , FiO ₂ , the use of mechanical ventilation. The MELD scoring system includes five indicators: serum bilirubin, creatinine (Scr), INR, and liver etiology or serum sodium. The formula is as follows: 3.8×ln[serum bilirubin (mg/dL)]+11.2×ln(INR)+9.6×ln[serum creatinine (mg/dL)]+6.4×(cause: cholestasis or alcohol is 0, others are 1). The MELD score is a classic model for the prognosis assessment of various liver diseases. The higher the score, the worse the prognosis. Patients with MELD>12 points can be included in the liver transplant candidate list. At the same time, according to the 2023 American Gastroenterological Association guidelines, the MELD score is recommended as a prognostic evaluation model for ALF patients. Since then, there have been continuous studies to improve MELD, deriving the MELD-Na, iMELD, and MESO scoring systems. The formula is as follows: MELD-Na = MELD + 1.59 × (135-Na, mmol/L), where Na>135mmol/L is calculated as 135mmol/L, Na<120mmol/L is calculated as 120mmol/L, and N a concentration of 120-135mmol/L is calculated according to the specific value; iMELD = MELD + 0.3 × age - 0.7 × Na + 100; MESO = [MELD / Na (mmol/L)] × 10. In 2015, Roayaie et al. proposed the PALBI score, which was originally used to determine the prognosis of liver cancer. In recent years, many studies have shown that the PALBI score has predictive power for the prognosis of liver cirrhosis, and is even better than MELD. The formula is as follows: (2.02 × log10 bilirubin) + [-0.37 × (log ABIC score includes age, total bilirubin, creatinine, and INR. It was originally developed to evaluate the prognosis of patients with alcoholic hepatitis. Existing studies have attempted to apply it to predict the short-term prognosis of patients with acute-on-chronic liver failure. The formula is as follows: (0.1×age)+[0.08×bilirubin (mg/dL)]+[0.3×creatinine (mg/dL)]+(0.8×INR). The TPPM score is a scoring system developed for HBV-related acute-on-chronic liver failure. The formula is as follows: P = 1/(1+e–logit(P)), logit(P) = 0.003×[total bilirubin (umol/L)]+0.951×INR+2.258×(complications: ≤1 complication is 0, ≥2 complications is 1)+0.114×[lg HBV DNA (copies/ml)]-5.012. None of these models can predict the prognosis of all liver failures. They mainly include indicators such as total bilirubin, INR, creatinine, and hepatic encephalopathy. Among them, the grading of hepatic encephalopathy is more subjective, and creatinine is easily affected by non-liver disease factors. Therefore, we need to select indicators with strong objectivity and fewer interference factors to construct a new scoring system.

At present, there is no definite model to predict the prognosis of all patients with liver failure. Therefore, this study aims to construct a simple score suitable for the prognosis of all patients with liver failure. The present invention selects four indicators: age, lactate, total bilirubin, and INR. These indicators are more objective, accurate, and simple, reflecting liver function and coagulation function. Many studies have shown that lactate is correlated with infection, hepatic encephalopathy, and body immunity in patients with liver failure. Strong inflammatory responses can lead to impaired mitochondrial oxidation function, increased anaerobic glycolysis, and the production of large amounts of lactic acid. In addition, elevated blood ammonia during hepatic encephalopathy increases glycolysis by stimulating phosphofructokinase, or inhibits α-ketoglutarate dehydrogenase, a key enzyme in the tricarboxylic acid cycle, interfering with brain energy metabolism, and leading to increased lactate formation. These accumulated brain lactic acids are likely to cause cerebral edema and increase the death rate in patients with liver failure. Therefore, these four independent prognostic factors reflect the pathophysiology of patients with liver failure.

The present invention shows that the ability of the LATI score in the development cohort to predict the 28-day and 90-day mortality of patients with liver failure is stronger than that of ABIC, CLIF-COF, AARC, iMELD, MESO, MELD, MELD-Na, and PALBI scores, and is comparable to the CLIF-CACLF score, which is a complex scale based on 6 organ failures and 11 predictive factors, including: liver: total bilirubin; kidney: creatinine and renal replacement therapy; nerve: hepatic encephalopathy grade; coagulation function: INR; circulation: mean arterial pressure and application of vasoactive drugs; respiration: PaO ₂ , SpO ₂ , FiO ₂ , and use of mechanical ventilation. In the internal and external validation cohorts, the LATI score also showed excellent discrimination and calibration. We further explored the application value of the prognostic model of the present invention in different types or different causes of liver failure, and the prediction effect of the LATI score was still satisfactory. It is worth noting that in the development cohort, the predictive value of the LATI score was higher than that of the AARC score, whether in all patients with liver failure or only in patients with acute-on-chronic liver failure; the predictive value of the LATI score was higher than that of the TPPM and MELD scores, whether in all patients with liver failure or only in patients with HBV-related acute-on-chronic liver failure.

The present invention divides patients with liver failure into high-risk group and low-risk group by the optimal cutoff value of LATI score. The results also show that the LATI score prognostic model of the present invention is simple and accurate in predicting the severity of patients with liver failure. In summary, the LATI score based on 4 indicators is an objective and reliable prediction model for patients with all types of liver failure. The model is more practical than ABIC, CLIF-C OF, AARC, iMELD, MESO, MELD, MELD-Na, PALBI, CLIF-C ACLF scores, and can accurately and timely judge the 28-day and 90-day prognosis of patients with liver failure and guide clinical treatment.

Claims (10)

1. A LATI prognostic model for patients with liver failure, characterized in that the model comprises four evaluation indicators: age, total bilirubin, INR and lactate, and the evaluation indicators are formed into a scoring system for predicting the 28-day mortality risk and 90-day mortality risk of patients with liver failure. 2. The prognostic model as described in claim 1, wherein the scoring system for predicting the 28-day mortality risk of patients with liver failure comprises a calculation formula: 28-day score = -7.04 + 0.21*lactate + 0.04*age + 0.01*total bilirubin + 0.69*INR. 3. The prognostic model of claim 1, wherein the scoring system for predicting the 90-day mortality risk of patients with liver failure comprises a calculation formula: 90-day score = -7.12 + 0.16*lactate + 0.05*age + 0.01*total bilirubin + 0.76*INR. 4. A method for constructing a LATI prognostic model according to any one of claims 1 to 3, comprising the following steps: 1) Recruit a sufficient sample size of patients with liver failure as a development cohort group; 2) Follow up the patients in the development cohort group for 28 or 90 days and collect the patient's index data, including age, general information, biochemical indicators, imaging examinations and complications; 3) The indicator data of the death group and the survival group at 28 days or 90 days in the development cohort group were analyzed by univariate and multivariate logistic regression to screen out the prognostic factors with significance or influence; 4) Based on the R "caret" software package, the indicator data of one-fifth of the patients in the development cohort were randomly selected as the objects of the training set of the machine learning algorithm, and the patient indicator data were used to construct the training set. The remaining people in the development cohort were assigned to the internal validation set, and the predictive performance of the factors affecting the training results was evaluated. The Logstic regression analysis algorithm was selected for further model construction; 5) Based on the results of logistic regression analysis, age, total bilirubin, lactate, and INR were determined to be independent influencing factors for the prognosis of patients with liver failure and used to construct a prediction model; 6) Construct the 28-day prediction model formula: 28-day LATI score = -7.04 + 0.21*lactate + 0.04*age + 0.01*total bilirubin + 0.69*INR, and construct the 90-day prediction model formula: 30-day LATI score = -7.12 + 0.16*lactate + 0.05*age + 0.01*total bilirubin + 0.76*INR. 5. The construction method according to claim 4, wherein the liver failure patients in step 1) include patients with HBV-related liver failure, patients with acute liver failure, patients with chronic liver failure or patients with chronic acute liver failure. 6. The construction method as described in claim 4, wherein the sample size of liver failure patients in step 1) is 1810 cases. 7. The construction method as claimed in claim 4, wherein the Logstic regression analysis score in step 4) further comprises filling the randomly missing data with a multiple interpolation method if the missing data exceeds 10% but does not exceed 50%. 8. The construction method according to claim 4, wherein the scoring in step 6) is to divide the 28-day and 90-day LAT I scores of patients with liver failure into a high-risk group and a low-risk group according to the cutoff values of the Kaplan-Meier survival curve shown in FIG6 . 9. Use of the LAT I prognostic model of claims 1 to 3 for evaluating and predicting the prognosis of all types of liver failure patients within 28 days or 90 days. 10. The use according to claim 9, wherein all types of liver failure patients include HBV-related liver failure patients, acute liver failure patients, chronic liver failure patients or chronic acute liver failure patients.