Construction method and system of acute obstructive suppurative cholangitis conservative treatment failure risk prediction model

Abstract

The invention provides a construction method and system of an acute obstructive suppurative cholangitis conservative treatment failure risk prediction model. The construction method comprises the following steps: S1, collecting clinical and laboratory index data of a plurality of cases of acute obstructive suppurative cholangitis patients conforming to a containing and discharging standard, preprocessing the clinical and laboratory index data, and then dividing the clinical and laboratory index data into a training set and a test set; s2, for the data of the training set, balancing is carried out by adopting an ROSE algorithm, and then feature screening is carried out by sequentially utilizing Spearman correlation analysis, single-factor ROC analysis and LASSO regression; s3, taking whether the patient is aggravated or not within 24 hours as a dependent variable, taking the features as independent variables, establishing a prediction model by adopting multivariable logistic regression, and optimizing model parameters through 10-fold cross validation; and S4, performing model performance verification evaluation based on the test set. The early prediction precision is significantly improved, and risk layering and treatment decision quantification are realized.

Description

Technical Field

[0001] This invention relates to the field of model building technology, and in particular to a method and system for constructing a risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis. Background Technology

[0002] Acute obstructive suppurative cholangitis (AOSC) is a severe acute abdomen caused by biliary obstruction complicated by bacterial infection. If not treated with timely drainage or surgery, it often rapidly progresses to sepsis and even multiple organ failure, with an extremely high mortality rate. The continuous updates to the Tokyo Guidelines (TG07, TG13, TG18) have provided international standards for the diagnosis and classification of AOSC. However, the guidelines lack specific criteria for the quantitative assessment of initial treatment failure in Grade II patients, only stating that "if there is no response to initial treatment, drainage or surgery should be performed as soon as possible." This vague definition makes it difficult for clinicians to accurately assess the trend of a patient's condition upon admission.

[0003] In large tertiary hospitals, teams experienced in ERCP or laparoscopic common bile duct exploration (LCBDE) can perform minimally invasive biliary drainage or stone removal at any time. However, in most primary hospitals, due to limitations in equipment and staff training, the number of hepatobiliary surgeons proficient in minimally invasive techniques is limited. In emergencies, they often have to provide anti-infection, fluid resuscitation, and other conservative treatments first. Assessing whether a patient's condition will worsen within 24 hours of admission is crucial in determining whether immediate transfer or emergency intervention is necessary. Identifying patients for whom conservative treatment is likely to succeed not only allows sufficient time to schedule elective minimally invasive treatment by experienced specialists but also significantly improves patient satisfaction, shortens hospital stays, and reduces treatment costs. Conversely, early identification of high-risk patients can prevent sepsis progression due to delays.

[0004] In summary, there is an uneven distribution of medical resources and limited emergency minimally invasive treatment capabilities in primary hospitals.

[0005] Common methods currently used to assess early changes in AOSC include:

[0006] Traditional clinical symptoms include Charcot's triad (fever, jaundice, and right upper quadrant pain), but its sensitivity is only about 50-70%, which is insufficient to identify high-risk patients in the early stages. .

[0007] The Tokyo Guideline (TG18) grading system relies on indicators such as body temperature, white blood cell count, and bilirubin for stratification, but the "no response to initial treatment" criterion lacks quantitative standards, making it difficult to make rapid decisions upon admission.

[0008] Single laboratory indicators, such as PCT, CRP, and ALT, can reflect inflammation, but their predictive efficacy is limited and they have not been used to form a composite discriminant model.

[0009] The accuracy of existing models is insufficient: A retrospective analysis of AOSC patients at Tongji Hospital over the past 10 years found that TG18 had a ROC AUC of only about 0.74 in predicting the worsening of the condition within 24 hours, which has limited clinical discrimination.

[0010] Furthermore, previous studies have mostly focused on treatment strategies for patients with severe (Grade III) AOSC, while systematic models for predicting the early course of the disease in patients with moderate AOSC are lacking. Especially in primary healthcare institutions, blindly delaying drainage may lead to a rapid deterioration of the condition, while premature surgery increases unnecessary trauma and costs, making it difficult to achieve precise treatment.

[0011] 1.Wada K, Takada T, Kawarada Y, et al. Diagnostic criteria and severity assessment of acute cholangitis: Tokyo Guidelines. J HepatobiliaryPancreat Surg. 2007;14(1):52-8. doi:10.1007 / s00534-006-1156-7.

[0012] 2.Kiriyama S, Takada T, Strasberg SM, et al. TG13 guidelines fordiagnosis and severity grading of acute cholangitis (with videos). JHepatobiliary Pancreat Sci. Jan 2013;20(1):24-34. doi:10.1007 / s00534-012-0561-3.

[0013] 3.Kiriyama S, Kozaka K, Takada T, et al. Tokyo Guidelines 2018:diagnostic criteria and severity grading of acute cholangitis (with videos).J Hepatobiliary Pancreat Sci. Jan 2018;25(1):17-30. doi:10.1002 / jhbp.512.

[0014] 4.Yokoe M, Takada T, Mayumi T, et al. Accuracy of the TokyoGuidelines for the diagnosis of acute cholangitis and cholecystitis takinginto consideration the clinical practice pattern in Japan. J HepatobiliaryPancreat Sci. Mar 2011;18(2):250-7. doi:10.1007 / s00534-010-0338-5.

[0015] 5.Kiriyama S, Takada T, Strasberg SM, et al. New diagnostic criteriaand severity assessment of acute cholangitis in revised Tokyo Guidelines. JHepatobiliary Pancreat Sci. Sep 2012;19(5):548-56. doi:10.1007 / s00534-012-0537-3.

[0016] 6.Tsuyuguchi T, Sugiyama H, Sakai Y, et al. Prognostic factors ofacute cholangitis in cases managed using the Tokyo Guidelines. JHepatobiliary Pancreat Sci. Sep 2012;19(5):557-65. doi:10.1007 / s00534-012-0538-2.

[0017] 7. Charcot M. On symptomatic hepatic fever - Comparison with uroseptic fever. Lessons on diseases of the liver, biliary tract and kidneys. 1877:176-185.

[0018] 8. Csendes A, Diaz JC, Burdiles P, Maluenda F, Morales E. Risk factors and classification of acute suppurative cholangitis. Br J Surg. Jul 1992;79(7):655-8. doi:10.1002 / bjs.1800790720.

[0019] 9. Lai EC, Tam PC, Paterson IA, et al. Emergency surgery for severe acute cholangitis. The high-risk patients. Ann Surg. Jan 1990;211(1):55-9.doi:10.1097 / 00000658-199001000-00009.

[0020] 10. Thompson JE, Jr., Tompkins RK, Longmire WP, Jr. Factors in management of acute cholangitis. Ann Surg. Feb 1982;195(2):137-45. doi:10.1097 / 00000658-198202000-00003.

[0021] 11.Haupert AP, Carey LC, Evans WE, Ellison EH. Acute suppurativecholangitis. Experience with 15 consecutive cases. Arch Surg. Apr 1967;94(4):460-8. doi:10.1001 / archsurg.1967.01330100024004. Summary of the Invention

[0022] To address the aforementioned technical problems, the first aspect of this application provides a method for constructing a risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis, comprising:

[0023] Step S1: Collect clinical and laboratory data of several patients with acute obstructive suppurative cholangitis who meet the inclusion and exclusion criteria, preprocess them, and then divide them into training set and test set;

[0024] Step S2: For the training set data, the ROSE algorithm is first used to balance the data, and then Spearman correlation analysis, univariate ROC analysis and LASSO regression are used in sequence to screen features. The final screened features include TBIL, LDH, IL-1β, IL-18, stone diameter and common bile duct diameter.

[0025] Step S3: Using whether the patient's condition worsens within 24 hours as the dependent variable and the feature as the independent variable, a multivariate logistic regression model is established. The model parameters are then optimized using 10-fold cross-validation. The model formula is as follows:

[0026] ;

[0027] Where P represents the probability that the patient's condition will worsen within 24 hours;

[0028] Step S4: Validate and evaluate the model performance based on the test set.

[0029] Furthermore, in step S3, the model formula is:

[0030] ;

[0031] The units for TBIL are μmol / L; LDH is U / L; IL-1β is pg / mL; IL-18 is pg / mL; the stone diameter is the largest stone diameter in mm; and the CBD diameter is mm.

[0032] Furthermore, in step S3, when P ≥ 0.23, the patient is deemed to have a high risk of conservative treatment failure within 24 hours.

[0033] The method for constructing the above-mentioned model for predicting the risk of conservative treatment failure in acute obstructive suppurative cholangitis further includes:

[0034] Step S5: Construct a nomogram: The nomogram is used to further verify the predictive conclusions of the risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis.

[0035] The second aspect of this application provides a risk prediction system for conservative treatment failure in acute obstructive suppurative cholangitis, comprising:

[0036] The indicator value input module is used to obtain the values ​​of TBIL, LDH, IL-1β, IL-18, stone diameter and CBD diameter of the current patient;

[0037] The model prediction module is used to perform probability prediction on the risk prediction model of conservative treatment failure in acute obstructive purulent cholangitis obtained by the above construction method.

[0038] The results output module is used to output the probability value of the patient's condition worsening within 24 hours, the corresponding risk of failure of conservative treatment within 24 hours, and the recommended treatment strategy.

[0039] The risk prediction system for conservative treatment failure in acute obstructive suppurative cholangitis further includes: a nomogram module, which automatically calculates the total score and the corresponding probability value of the current patient's condition worsening within 24 hours based on the values ​​of TBIL, LDH, IL-1β, IL-18, stone diameter and CBD diameter of the current patient and the nomogram.

[0040] By adopting the above technical solution, this application has the following beneficial technical effects:

[0041] This invention, based on the existing Tokyo Guidelines (TG18) and traditional clinical assessment systems, proposes an early risk prediction model based on multi-dimensional indicator integration. This model enables individualized risk stratification of patients with Grade II acute obstructive suppurative cholangitis (AOSC) upon admission. The model not only significantly outperforms existing standards in predictive performance but also demonstrates good scalability and clinical operability. This invention fills the gap in early dynamic risk assessment within the TG18 grading system, and is particularly suitable for the real-world scenario where minimally invasive techniques are not yet widely available in primary hospitals. It provides an operable, scalable, and quantifiable scientific tool for the tiered diagnosis and treatment of AOSC patients, offering decision support for clinicians at different levels of hospitals. Its beneficial effects are reflected in the following aspects:

[0042] (i) Significantly improve the accuracy of early predictions

[0043] The existing TG18 system relies on subjective experience and a single indicator to judge the changes in the condition of patients with moderate AOSC, lacks quantitative standards, and has limited predictive performance (AUC ≈ 0.74).

[0044] This invention integrates six independent predictors—total bilirubin (TBIL), lactate dehydrogenase (LDH), interleukin-1β (IL-1β), interleukin-18 (IL-18), common bile duct stone diameter, and common bile duct diameter—through multivariate logistic regression to construct a composite model, which significantly improves the discriminative efficacy.

[0045] On the independent test set, the model achieved an AUC of 0.937 (95% CI: 0.890–0.984), a sensitivity of 94.1%, and a specificity of 80.1%, representing an improvement of approximately 20% compared to the TG18 system.

[0046] This high-precision predictive ability enables clinicians to identify potential risks of disease deterioration within the first 24 hours of a patient's admission, thus avoiding delays in treatment.

[0047] (ii) Achieving risk stratification and quantification of treatment decisions

[0048] By transforming the regression equation into a nomogram, this invention visualizes the abstract statistical model, allowing doctors to directly calculate the total score based on six indicators and convert it into an individualized risk probability.

[0049] The optimal threshold of 0.23 (maximizing the Youden Index) was determined by ROC analysis of the model, corresponding to a total score of approximately 180 points on the nomogram.

[0050] A predicted probability ≥ 0.23 indicates a high risk of conservative treatment failure within 24 hours, and ERCP or surgical drainage should be performed as soon as possible;

[0051] If the predicted probability is < 0.23, it indicates that conservative treatment is safe and feasible, and observation can continue.

[0052] This quantitative stratification mechanism transforms the vague "no response to initial treatment" in TG18 into a clear numerical standard for the first time, enabling clinical decision-making to shift from experience-based judgment to data-driven objective reasoning, and achieving quantitative standardization of early management of cholangitis.

[0053] (III) Multidimensional integrated indicators to reflect the whole picture of the disease

[0054] Traditional diagnostic methods often rely on single biochemical or clinical indicators, while this invention integrates inflammatory factors, metabolic indicators, and imaging parameters to reflect different physiological aspects of disease progression.

[0055] TBIL and LDH reveal the degree of biliary obstruction and hepatocellular damage;

[0056] IL-1β and IL-18 reflect the intensity of the host's inflammatory response and can capture early signals of sepsis progression;

[0057] The diameter of the gallstone and the diameter of the common bile duct quantify the degree of mechanical obstruction.

[0058] This multidimensional information fusion enables the model to comprehensively reflect pathophysiological changes, making it more stable and generalizable than traditional single-indicator models.

[0059] (iv) The model is robust and reliable, and has clinical applicability.

[0060] After 200 internal bootstrap iterations and independent external validation, the calibration curve closely matches the ideal curve, with a Brier score of 0.067 and a mean absolute error of 0.008, indicating that the prediction results are stable and reliable.

[0061] All the required indicators come from routine laboratory tests and imaging examinations, which can be completed in primary hospitals without the need for high-end testing equipment or additional costs.

[0062] The model's algorithm has a simple structure, relying on only six core variables. It can be embedded in electronic medical record systems for automatic calculation, or quickly interpreted in a computer-free environment using paper nomograms, facilitating widespread application by clinicians. It assists primary care physicians in making rapid decisions: the model can be applied through nomograms or simple software to help primary care physicians, even with limited experience, quantitatively determine whether a patient can continue conservative treatment or should be transferred to a higher-level hospital for minimally invasive drainage as soon as possible.

[0063] (V) Optimize the allocation of medical resources and patient benefits

[0064] This invention is particularly suitable for primary care and non-specialized hospitals, solving the practical problems of insufficient mastery of minimally invasive techniques and limited emergency treatment capabilities.

[0065] By immediately stratifying patients upon admission, doctors can refer high-risk patients in advance or initiate drainage preparations, thereby reducing the incidence of acute sepsis and organ failure.

[0066] At the same time, for low-risk patients, blind early surgery can be avoided, saving medical resources, shortening hospital stay, reducing treatment costs, and significantly improving patient satisfaction and overall treatment efficiency.

[0067] This mechanism has significant value for promotion in areas with limited medical resources and can serve as a decision-making tool for the management of acute cholangitis within a tiered healthcare system.

[0068] Optimizing resources and time: Through scientific stratification, low-risk patients can safely receive 24-hour conservative treatment and elective minimally invasive procedures, while high-risk patients receive timely intervention, significantly improving overall treatment efficiency.

[0069] Enhancing patient benefits: Achieving individualized management that prioritizes conservative approaches and provides early intervention for urgent needs, thereby improving patient satisfaction, shortening hospital stays, and reducing medical expenses.

[0070] (vi) It has the potential for expansion and substitution.

[0071] This modular design allows the invention to serve not only AOSC patients but also to be extended to early warning of other infectious biliary tract diseases.

[0072] In summary, this invention, with a six-factor integrated model at its core, achieves high-precision prediction, risk quantification, and decision visualization of AOSC patients' condition worsening within 24 hours. It has significant clinical and social value in improving diagnosis and treatment efficiency, reducing overtreatment, optimizing resource allocation, and improving patient prognosis. Attached Figure Description

[0073] Figure 1 Flowchart for patient screening and analysis.

[0074] Figure 2This section presents the screening of candidate predictive factors and the discriminant power of single indicators in patients with TG18-grade moderate (Grade II) acute obstructive suppurative cholangitis. Figure a shows the Spearman correlation heatmap of candidate variables, illustrating the correlation between various clinical, biochemical, and inflammatory factors. Variables with |ρ| ≥ 0.4 are highlighted. Total bilirubin (TBIL), direct bilirubin (DBIL), interleukin-1β (IL-1β), interleukin-18 (IL-18), and common bile duct stone diameter showed significant correlations with multiple indicators, suggesting potential core predictive value. Figure b shows the univariate ROC curve analysis. Due to multicollinearity between TBIL and DBIL, the discriminant power of nine candidate variables (IL-18, IL-1β, TBIL, LDH, CBD stone diameter, CBD diameter, IFN-γ, PCT, and IL-6) was compared after removing DBIL. The results showed that IL-18 (AUC = 0.852) and CBD stone diameter (AUC = 0.841) had the highest predictive performance, followed by TBIL and IL-1β; CBD diameter and LDH showed moderate performance, while IFN-γ, PCT, and IL-6 had limited discriminative ability. Figure c shows the LASSO regression coefficient path. As log(λ) increases, the variable coefficients gradually shrink to near zero, showing the stability and sparsity of the variable selection process. Figure d shows the λ selection curve under 10-fold cross-validation. It illustrates the relationship between binomial bias and log(λ), and marks the positions of λ_min and λ_1SE. The final model retains six non-zero variables at λ_1SE: TBIL, LDH, IL-1β, IL-18, CBD stone diameter, and CBD diameter, forming the final multivariate prediction model.

[0075] Figure 3 The results of establishing and validating a multivariate logistic regression model for the initial 24-hour conservative treatment of patients with TG18 grade II acute obstructive purulent cholangitis. Figure a shows the ROC curves of the logistic regression model after ROSE equilibrium on the training set (blue solid line) and the independent test set (red dashed line). The curves show the classification performance of the model on the two sets of data: AUC of 0.951 (95% CI: 0.937–0.966) on the training set and AUC of 0.937 (95% CI: 0.890–0.984) on the test set. The dots on the curve represent the Youden optimal threshold (t=0.564 on the training set and t=0.235 on the test set). The model exhibits high sensitivity and good discriminative ability. Figure b shows the Bootstrap internal calibration curve (B=200) on the training set. It shows the comparison of the apparent curve, bias correction curve, and ideal 45° line. The mean absolute error is 0.008, and the mean square error is 9×10⁻⁻⁻⁴.5 The 90th percentile absolute error is 0.014, indicating that the model's predicted probabilities match the actual observed events well. Figure c shows the calibration results on the test set. The Brier score in external validation is 0.067 with a slope of 1.078, and the overall calibration value is −1.572; Spiegelhalter's Z = −2.19 (p = 0.029), suggesting a slight prediction bias but good overall fit. Figure d shows the paired ROC comparison analysis between the model and the TG18 classification system. The nomograph model has an AUC of 0.937 (95% CI: 0.890–0.984), which is significantly better than the logistic model based on TG18 (AUC 0.746, 95% CI: 0.596–0.897). The difference between the two is statistically significant according to the DeLong test (ΔAUC=0.191, p=0.0025).

[0076] Figure 4 A nomogram to predict the 24-hour success rate of conservative treatment in patients with acute obstructive suppurative cholangitis. Detailed Implementation

[0077] The advantages of the present invention are further illustrated below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that the following detailed description is illustrative rather than restrictive and should not be construed as limiting the scope of protection of the present invention.

[0078] Example 1: Method for constructing a risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis

[0079] The method for constructing a risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis provided in this embodiment includes steps S1-S4:

[0080] Step S1: Collect clinical and laboratory data of several patients with acute obstructive suppurative cholangitis who meet the inclusion and exclusion criteria, preprocess them, and then divide them into training and test sets.

[0081] The data for this invention comes from patients with acute obstructive suppurative cholangitis admitted to Tongji Hospital, affiliated with Tongji University, between January 1, 2013, and December 31, 2023. All cases were radiographically confirmed to have biliary obstruction and met the diagnostic and grading criteria of the Tokyo Guidelines 2018 (TG18). The overall research process includes patient screening, data collection, feature screening, model construction, validation, and clinical application. Figure 1 As shown.

[0082] (a) Patient screening and data sources

[0083] Inclusion criteria were: (1) meeting the diagnostic criteria for TG18 moderate (Grade II) AOSC; (2) stable vital signs and no signs of shock upon admission; (3) having any two of the following: abnormal white blood cell count (>12,000 / μL or <4,000 / μL), body temperature >38.5℃, age ≥75 years, total bilirubin ≥85.5 μmol / L (5 mg / dL), serum total protein below 70% of the reference limit; (4) imaging showing dilatation of the common bile duct or visible stones; (5) the patient and family consent to at least 24 hours of conservative treatment. Exclusion criteria were: (1) shock or death upon admission; (2) concurrent failure of other organ functions (such as heart, lung, and kidney failure); (3) immediate emergency surgery or ERCP drainage upon admission; (4) missing clinical data or incomplete key variables.

[0084] A total of 964 patients met the inclusion criteria for the analysis. Of these, 880 patients remained stable after 24 hours of conservative treatment and were defined as the control group (Moderate AOSC group); 84 patients experienced clinical deterioration within 24 hours, such as high fever, shock, or a rapid increase in bilirubin, requiring emergency surgery or ERCP and were defined as the experimental group (Moderate with Aggravation group). There were no statistically significant differences between the two groups in baseline characteristics such as sex, age, body mass index, and prevalence of diabetes. The primary endpoint of the study was the incidence of exacerbations within 24 hours.

[0085] (II) Collection of Clinical and Laboratory Indicators

[0086] All patients underwent routine physical examinations, venous blood sample collection, and abdominal imaging assessments within 2 hours of admission.

[0087] Clinical indicators included vital signs such as body temperature, systolic blood pressure, and diastolic blood pressure; biochemical indicators included total bilirubin (TBIL), direct bilirubin (DBIL), lactate dehydrogenase (LDH), alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), alkaline phosphatase (ALP), blood urea nitrogen (BUN), and creatinine (Cr), all of which were measured using a fully automated biochemical analyzer (Roche Cobas series); inflammatory factor indicators included interleukin-1β (IL-1β), interleukin-6 (IL-6), interleukin-18 (IL-18), interferon-γ (IFN-γ), procalcitonin (PCT), and C-reactive protein (CRP), which were detected using multiplex immunoturbidimetry or enzyme-linked immunosorbent assay (ELISA); imaging parameters included the diameter of the common bile duct stone and the diameter of the common bile duct (CBD), which were independently measured by two radiologists based on abdominal CT images and the average value was taken.

[0088] All tests were conducted in accordance with the hospital's standard operating procedures (SOPs), and the results were entered into the database after being reviewed and confirmed by the laboratory's quality control department.

[0089] (III) Data Processing and Feature Filtering

[0090] To ensure model stability, variables with a missing rate exceeding 20% ​​were first removed, and the remaining data were subjected to multiple imputation and Z-score standardization.

[0091] It was then divided into training and testing sets.

[0092] Step S2: For the training set data, first use the ROSE algorithm to balance the data, and then use Spearman correlation analysis, univariate ROC analysis and LASSO regression to select features in sequence.

[0093] Since the two groups of samples are severely unevenly distributed (the additive group accounts for 8.7%), the ROSE (Random Over-SamplingExamples) algorithm is used to balance the training set in order to improve the model's performance in recognizing minority class samples.

[0094] Subsequently, Spearman correlation analysis was used to screen variables with correlation coefficients |ρ| ≥ 0.4, eliminating multicollinearity factors; then, one-way ROC analysis was performed to compare the predictive power of each indicator. The results showed that IL-18 (AUC = 0.852), stone diameter (AUC = 0.841), and TBIL (AUC = 0.818) had the highest discriminant power (see...). Figure 2 (Variable selection). Based on the LASSO regression selection results, six indicators were finally selected for multivariate modeling: TBIL, LDH, IL-1β, IL-18, stone diameter, and common bile duct diameter.

[0095] Step S3: Using whether the patient's condition worsens within 24 hours as the dependent variable and the feature as the independent variable, a prediction model is established using multivariate logistic regression and the model parameters are optimized through 10-fold cross-validation.

[0096] Using whether the patient's condition worsened within 24 hours as the dependent variable (1 = worsening, 0 = stable), and six core indicators as independent variables, a multivariate logistic regression predictive model was established. The model format is as follows:

[0097]

[0098] Where P represents the probability that the patient's condition will worsen within 24 hours. When P ≥ 0.23, the patient is considered to have a high risk of failure of conservative treatment within 24 hours.

[0099] Preferably, the model formula is:

[0100] ;

[0101] The units for TBIL are μmol / L; LDH is U / L; IL-1β is pg / mL; IL-18 is pg / mL; the stone diameter is the largest stone diameter in mm; and the CBD diameter is mm.

[0102] Model parameters were optimized using 10-fold cross-validation.

[0103] Step S4: Validate and evaluate the model performance based on the test set.

[0104] Performance was evaluated on an independent test set. Results showed that the model's ROC curve AUC was 0.937 (95% CI: 0.890–0.984), with a sensitivity of 94.1% and a specificity of 80.1%, significantly superior to the TG18 system (AUC 0.746, ΔAUC = 0.191, p = 0.0025) (see [link to relevant documentation]). Figure 3 Model performance comparison).

[0105] After 200 internal validations using the bootstrap method, the model achieved a Brier score of 0.067 and a mean absolute error of 0.008. The calibration curve fits the ideal line well, indicating that the prediction results are stable and reliable.

[0106] Example 2: Risk Prediction System for Conservative Treatment Failure in Acute Obstructive Suppurative Cholangitis

[0107] Based on the risk prediction model for conservative treatment failure of acute obstructive suppurative cholangitis already constructed in Example 1, a risk prediction system for conservative treatment failure of acute obstructive suppurative cholangitis is further constructed, including: an indicator value input module, a model prediction module, a nomogram module, and a result output module.

[0108] The indicator value input module is used to obtain the values ​​of TBIL, LDH, IL-1β, IL-18, stone diameter and CBD diameter of the current patient.

[0109] Patient Basic Information and Admission Status: The patient, a 68-year-old female, was admitted to the emergency department due to "fever, jaundice, and upper abdominal pain for 12 hours." She had a history of cholelithiasis but no history of diabetes or cardiac or renal dysfunction. On admission, her temperature was 38.6℃, blood pressure was 115 / 75 mmHg, and pulse was 92 beats / min. Physical examination revealed mild tenderness in the right upper quadrant, and Murphy's sign was negative.

[0110] Laboratory and Imaging Examinations: The following tests were completed within 2 hours of admission: Complete Blood Count: White blood cell count 14,800 / μL (↑); Biochemical Indicators: Total bilirubin (TBIL) = 118.2 μmol / L, Direct bilirubin (DBIL) = 60.5 μmol / L, Lactate dehydrogenase (LDH) = 389 U / L, Alanine aminotransferase (ALT) = 108 U / L, Aspartate aminotransferase (AST) = 121 U / L. Inflammatory Factor Detection (ELISA): IL-1β = 19.4 pg / mL, IL-18 = 254.6 pg / mL. Imaging Examination (Upper Abdominal CT, double-checked by two physicians): Common bile duct (CBD) diameter = 8.9 mm; Maximum stone diameter = 8.1 mm.

[0111] According to the TG18 criteria, the patient meets the diagnostic criteria for Grade II acute obstructive suppurative cholangitis. The physician plans to provide 24-hour conservative treatment (broad-spectrum antibiotics, fluid resuscitation, and monitoring).

[0112] The model prediction module is used to predict the probability of failure of conservative treatment for acute obstructive suppurative cholangitis obtained by the construction method in Example 1.

[0113] Model Application: Variable Value Input: Input the patient's six core indicators into the prediction model.

[0114] TBIL=118.2, LDH=389, IL-1β=19.4, IL-18=254.6, stone diameter=8.1, common bile duct diameter=8.9.

[0115] Calculate the logistic regression linear predictor (Logit) according to the model formula (coefficients are virtual sample values):

[0116] Logit(P) = −19.69 + 0.037 × TBIL + 0.005 × LDH + 0.103 × IL−1β + 0.021 × IL−18 + 0.575 × stone diameter + 0.345 × CBD diameter

[0117] Substituting the values, we get: Logit(P) = −19.69 + 0.037(118.2) + 0.005(389) + 0.103(19.4) + 0.021(254.6) + 0.575(8.1) + 0.345(8.9) = 1.701. The converted prediction probability (P) is 0.846, which means that the predicted probability of the patient's condition worsening during the 24-hour conservative treatment period is 84.6%.

[0118] Risk assessment and clinical interpretation: Based on the optimal threshold determined by ROC analysis (cutoff = 0.23), this patient's predicted probability (0.846) > 0.23, classifying them as high-risk. This suggests a high risk of conservative treatment failure within 24 hours, and ERCP or surgical drainage should be performed as soon as possible.

[0119] The nomogram module is used to automatically calculate the total score and the corresponding probability value of the current patient's condition worsening within 24 hours based on the values ​​of TBIL, LDH, IL-1β, IL-18, stone diameter and CBD diameter of the current patient and the nomogram.

[0120] To facilitate bedside application, the model parameters are visualized as a nomogram, with each variable corresponding to a scale line. The system automatically reads and sums the scores based on the patient's indicator values ​​to obtain an individualized probability of exacerbation risk (see...). Figure 4 (Nomogram illustration). When the predicted probability exceeds 0.23 (i.e., the nomogram score is greater than 180), it indicates that the patient has a high risk of conservative treatment failure and ERCP or surgical drainage should be performed as soon as possible; otherwise, conservative observation can continue for 24 hours.

[0121] This nomogram features a simple structure and intuitive operation, and can be embedded in electronic medical record systems or mobile terminal software to achieve automated calculations. All the tests involved are routine clinical examinations, suitable for widespread use in hospitals at all levels. It is particularly helpful for primary care physicians to quickly make early risk assessments under conditions of limited minimally invasive equipment and human resources, thereby improving diagnostic and treatment efficiency, optimizing resource allocation, and improving patient prognosis.

[0122] Based on the nomogram (e.g.) Figure 4 Automatic reading:

[0123] TBIL=118 μmol / L corresponds to 52 points;

[0124] LDH=389 corresponds to 21 points;

[0125] IL-1β=19 corresponds to 27 points;

[0126] IL-18=255, corresponding to 55 points;

[0127] Stone diameter = 8.1 mm, corresponding to 42 points;

[0128] The common bile duct diameter is 8.9 mm, corresponding to 22 points.

[0129] The total score is 219 points.

[0130] According to the probability scale below the nomogram, a score of 219 corresponds to a predicted probability of 0.85, consistent with the logistic regression calculation result. Doctors can obtain the same conclusion intuitively in the ward without complex calculations.

[0131] The results output module is used to output the probability value of the patient's condition worsening within 24 hours, the corresponding risk of failure of conservative treatment within 24 hours, and the recommended treatment strategy.

[0132] Clinical Management and Outcome Verification: This patient initially received conservative treatment, but subsequently developed shock. Fortunately, timely surgery saved her life. This case validates the clinical value and accuracy of the model's predictions.

[0133] Through the above embodiments, those skilled in the art can quickly reproduce the calculation and judgment process of the model using routine biochemical tests and imaging measurement data, without additional experiments, following the same procedure. The indicators required by the model are all standardized hospital testing items, and the calculation formulas can be embedded in spreadsheets, medical record systems, or mobile applications, making them highly operable.

[0134] This method is not only applicable to individualized risk assessment of single patients, but can also be extended to batch prediction of multicenter cohorts to establish regional cholangitis early warning systems and provide decision support for primary hospitals.

[0135] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for constructing a risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis, characterized in that, include: Step S1: Collect clinical and laboratory data of several patients with acute obstructive suppurative cholangitis who meet the inclusion and exclusion criteria, preprocess them, and then divide them into training set and test set; Step S2: For the training set data, the ROSE algorithm is first used to balance the data, and then Spearman correlation analysis, univariate ROC analysis and LASSO regression are used in sequence to screen features. The final screened features include TBIL, LDH, IL-1β, IL-18, stone diameter and common bile duct diameter. Step S3: Using whether the patient's condition worsens within 24 hours as the dependent variable and the feature as the independent variable, a multivariate logistic regression model is established. The model parameters are then optimized using 10-fold cross-validation. The model formula is as follows: ； Where P represents the probability that the patient's condition will worsen within 24 hours; Step S4: Validate and evaluate the model performance based on the test set.

2. The method for constructing a risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis as described in claim 1, characterized in that, In step S3, the model formula is: ； The units for TBIL are μmol / L; LDH is U / L; IL-1β is pg / mL; IL-18 is pg / mL; the stone diameter is the largest stone diameter in mm; and the CBD diameter is mm.

3. The method for constructing a risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis as described in claim 2, characterized in that, In step S3, when P ≥ 0.23, the patient is considered to have a high risk of conservative treatment failure within 24 hours.

4. The method for constructing a risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis as described in claim 1, characterized in that, Further includes: Step S5: Construct a nomogram: The nomogram is used to further verify the predictive conclusions of the risk prediction model for conservative treatment failure in acute obstructive suppurative cholangitis.

5. A system for predicting the risk of conservative treatment failure in acute obstructive suppurative cholangitis, characterized in that: include: The indicator value input module is used to obtain the values ​​of TBIL, LDH, IL-1β, IL-18, stone diameter and CBD diameter of the current patient; The model prediction module is used to perform probability prediction using the risk prediction model for conservative treatment failure of acute obstructive purulent cholangitis obtained by the construction method as described in any one of claims 1 to 3. The results output module is used to output the probability value of the patient's condition worsening within 24 hours, the corresponding risk of failure of conservative treatment within 24 hours, and the recommended treatment strategy.

6. The risk prediction system for conservative treatment failure of acute obstructive suppurative cholangitis as described in claim 5, characterized in that, Further includes: The nomogram module is used to automatically calculate the total score and the corresponding probability value of the current patient's condition worsening within 24 hours based on the values ​​of TBIL, LDH, IL-1β, IL-18, stone diameter and CBD diameter of the current patient and the nomogram.

Images