DKA combined AP risk prediction model based on multi-index fusion, medium and equipment

Abstract

The invention discloses a DKA combined AP risk prediction model based on multi-index fusion, a medium and equipment, and relates to the technical field of medical health and artificial intelligence, and the risk prediction model comprises a preprocessing unit which is used for carrying out the standardization processing of basic data and clinical data of a diabetic ketoacidosis patient, the clinical data comprises stomachache symptoms, C reactive protein level and serum lipase level; the screening unit is used for screening the basic data according to an exclusion mechanism; the data integration unit is used for obtaining the probability of occurrence of DKA and AP events by combining the following formula according to the screening result: Q = 1 / (1 + e-(beta0 + beta1 * stomachache + beta2 * CRP + beta3 * LPS)); and the output unit is used for outputting the probability calculated by the data integration unit as a result. Under the emergency treatment scene, the risk of acute pancreatitis complicated by the diabetic ketoacidosis patient can be rapidly evaluated, and a good preposed basis is provided for clinical decision making.

Description

Technical Field

[0001] The present invention relates to the fields of medical health and artificial intelligence technology, and in particular to a DKA combined with AP risk prediction model, medium and equipment based on multi-indicator fusion. Background Art

[0002] Diabetic ketoacidosis (DKA) is an acute complication of diabetes mellitus caused by a variety of factors. Insufficient insulin secretion and abnormally elevated glucose-raising hormones lead to disturbances in sugar, protein, and fat metabolism, as well as imbalances in water, electrolyte, and acid-base balance, resulting in symptoms such as hyperglycemia, hyperketosis, and metabolic acidosis. Previous studies have shown that patients with type 1 diabetes are prone to spontaneous ketoacidosis, while patients with type 2 diabetes can also develop ketoacidosis under certain predisposing factors. These common predisposing factors include infection, interruption or inappropriate reduction of insulin therapy, improper diet, and various stresses such as trauma, surgery, pregnancy, and childbirth. Acute pancreatitis (AP), a common clinical acute abdomen, is characterized by activation of pancreatic enzymes and a localized pancreatic inflammatory response. It can be caused by a variety of factors, including cholelithiasis, alcohol abuse, and pancreatic duct obstruction, and may be accompanied by a systemic inflammatory response. Since DKA and AP have a high degree of overlap in clinical manifestations (such as abdominal pain, nausea and vomiting) and laboratory indicators (such as increased serum amylase AMY and lipase LPS), it is clinically difficult to distinguish whether DKA patients have AP.

[0003] Traditional clinical differentiation mainly relies on a single indicator (such as an increase in AMY / LPS ≥3 times the upper limit of normal) or empirical judgment. For example, Literature 1: "Analysis of Clinical Characteristics of Patients with Diabetic Ketoacidosis and Acute Pancreatitis in Children and Adolescents" (Wei Liya et al., "Chinese Journal of Diabetes", Vol. 12, No. 4, 2020) proposed that in children with DKA, dynamic monitoring of pancreatic enzyme (amylase / lipase) changes and combined with imaging examinations (CT) to exclude stress-induced pancreatic enzyme elevation can reduce the misdiagnosis rate of AP; Literature 2: "Clinical Analysis of 13 Cases of Diabetic Ketoacidosis Combined with Hypertriglyceridemia and Acute Pancreatitis" (He Minghai et al., "Western Medicine", Vol. 21, No. 10, 2009) identified hypertriglyceridemia (≥11.3 mmol / L) as the core cause of acute pancreatitis (AP), and also proposed a combined diagnostic strategy combining pancreatic enzyme (amylase / LPS) detection with imaging.

[0004] However, the above method has obvious limitations: some DKA patients may experience elevated AMY / LPS due to stress (such as increased secretion of salivary AMY), dehydration (such as decreased clearance of AMY / LPS) or metabolic disorders (such as increased lipolysis caused by insulin deficiency), even if AP is not present. For example, Yadav D et al. found that 16% to 25% of DKA patients experienced nonspecific elevations of AMY and LPS, that is, some DKA patients experienced non-AP-related elevations of AMY and LPS (Yadav D, Nair S, Norkus EP, et at. Nonspecific hyperamylasemia and hyperlipasemia in diabeticketoacidosis: incidence and correlation with biochemical abnormalities [J]. Am J Gastroenterol, 2000, 95 (11)). Therefore, the combined diagnosis of pancreatic enzyme testing and imaging will result in significant lag, and the problem of rapid initial screening of the risk of AP in emergency patients with DKA is not solved.

[0005] In addition, literature 3: "Analysis of clinical characteristics and prognostic factors in patients with diabetic ketoacidosis combined with acute pancreatitis" (Yang Lijun, "China Science and Technology Journal Database Medicine", Issue 5, 2025) identified blood glucose > 21.75 mmol / L as a laboratory parameter for identifying high-risk patients through multivariate regression analysis; Literature 4: "Clinical characteristics of diabetic ketoacidosis combined with acute pancreatitis" (Chen Mengjie, Master's thesis of Dalian Medical University, 2023) identified abdominal pain, duration of abdominal pain, total cholesterol (TC) and triglycerides (TG) as independent risk factors for DKA combined with AP through multivariate logistic regression analysis.

[0006] However, the regression models in References 3 and 4 are complex and time-consuming to test. Reference 3 primarily focuses on the age and prognosis of patients with DKA and AP, and its regression model includes 19 independent variables (e.g., PaCO2, HCT, CRP, triglycerides, etc.); the regression model in Reference 4 includes 14 independent variables (e.g., body mass index, cholesterol, triglycerides, etc.). Furthermore, in References 3 and 4, metabolic markers (e.g., blood glucose, triglycerides, and TC) dominate and require confirmation with imaging studies. This is because DKA itself can lead to elevated triglycerides (i.e., insulin deficiency → increased lipolysis), which may induce pancreatitis (microcirculatory impairment) but does not directly reflect the degree of pancreatic tissue damage and, therefore, has no necessary causal relationship with AP. Consequently, References 3 and 4 also fail to address the issue of rapid initial screening for the risk of AP in emergency patients with DKA.

[0007] In summary, although some key diagnostic indicators for diabetic ketoacidosis (DKA) patients with acute pancreatitis (AP) have been continuously clarified, effectively reducing the misdiagnosis and missed diagnosis rates, the adaptability of these solutions in emergency scenarios is not ideal. The main reason is that for DKA patients, although existing solutions follow some diagnostic gold standards (such as metabolic testing and CT imaging), these solutions are time-consuming (generally exceeding 5 hours). In emergency scenarios, for hospitals, especially comprehensive hospitals that need to cope with the diagnosis needs of multiple diseases, and emergency patients, the dynamic supply and demand balance and medical resource scheduling issues are more obvious. Summary of the Invention

[0008] In response to the above-mentioned deficiencies in the prior art, the present invention provides a DKA combined with AP risk prediction model, medium and device based on multi-indicator fusion.

[0009] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0010] The risk prediction model for DKA combined with AP based on multi-indicator fusion includes:

[0011] a pretreatment unit for standardizing basic information and clinical data of patients with diabetic ketoacidosis, wherein the clinical data include abdominal pain symptoms, C-reactive protein levels, and serum lipase levels;

[0012] A screening unit, configured to screen the basic information according to an exclusion mechanism; the exclusion mechanism includes the basic information meeting any one or two conditions of type 1 diabetes, childhood diabetes, and adolescent diabetes;

[0013] The data integration unit is used to obtain the probability of DKA combined with AP events based on the screening results of the screening unit and the following formula:

[0014] Q = 1 / (1 + e^-(β0 + β1 × abdominal pain + β2 × CRP + β3 × LPS))

[0015] Where Q is the probability of DKA combined with AP; e is the base of the natural logarithm; β0 is the intercept term, whose value is obtained by maximum likelihood estimation; abdominal pain is a binary variable with values ​​of 1 and 0; CRP represents C-reactive protein data, in mg / L; LPS represents serum lipase data, in U / L; β1, β2, and β3 are the regression coefficients of abdominal pain, CRP, and LPS, respectively, and the values ​​of β1, β2, and β3 range from 1.225 to 4.688, 0.005 to 0.021, and 0.304 to 2.006, respectively.

[0016] The output unit is used to output the probability calculated by the data integration unit as a result.

[0017] Furthermore, the preprocessing unit also includes missing value processing and variable standardization, wherein the missing value processing uses the multiple interpolation method to generate several complete data sets and combines these data sets to fill the missing values, and the variable standardization uses the Z-score standardization method to eliminate the impact of dimensional differences on the model.

[0018] Preferably, the value of β0 is -12.406.

[0019] Preferably, the values ​​of β1, β2, and β3 are 2.348, 0.013, and 0.908, respectively.

[0020] Based on the above-mentioned risk prediction model, the present invention also provides a computer-readable storage medium storing computer instructions, which execute the functions of the above-mentioned risk prediction model when executed.

[0021] In addition, based on the above-mentioned risk prediction model, the present invention also provides a computer device, including a memory and a processor; the memory is used to store computer instructions; the processor runs the computer instructions to implement the functions of the above-mentioned risk prediction model.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) The present invention addresses the problems existing in the prior art. Through creative work, abdominal pain, CRP, and LPS are selected from a number of key diagnostic indicators as independent risk factors for DKA combined with AP. A three-dimensional fusion system of "abdominal pain + CRP + LPS" is set, and the values ​​of coefficients β1, β2, and β3 are designed, and abdominal pain is classified as a binary variable of 1 and 0, thereby constructing a multi-dimensional dynamic risk prediction model. The model can quickly output a quantitative risk probability (Q value), realize auxiliary initial screening of the risk of DKA combined with AP, meet the needs of rapid emergency triage, and better adapt to the dynamic supply and demand balance in emergency scenarios. In addition, the solution of the present invention can also help doctors determine whether further examinations are needed first. For example, when the risk prediction probability of DKA combined with AP is low, doctors can comprehensively consider whether to postpone imaging examinations based on actual conditions (including clinical judgment, current department staffing, etc.); conversely, when the risk prediction probability of DKA combined with AP is high, doctors can immediately arrange imaging examinations. This is consistent with the needs of emergency resource optimization and realizes the effective scheduling of medical resources. The above technical effects are due to the deep integration of the present invention in indicator screening, model construction and parameter value range design.

[0024] Furthermore, implementation of the present invention's approach can also avoid misdiagnosis or missed diagnosis due to nonspecific elevations of a single indicator. For example, some DKA patients may experience elevated AMY or LPS levels due to stress reactions or metabolic disorders, without concurrent AP. This invention effectively eliminates this interference through multi-indicator fusion analysis. Furthermore, the model independently outputs risk probabilities, providing a reasonable and feasible pre-treatment solution for early identification of AP in emergency DKA patients.

[0025] Furthermore, the present invention requires only three input indicators: abdominal pain, CRP, and LPS. Compared to metabolic screening and high triglyceride diagnosis, CRP and LPS are both routine emergency tests, and the test takes less than two hours, or even less. Combined with the binary classification of abdominal pain, this method can quickly complete the risk assessment of DKA emergency patients after admission. Furthermore, grassroots hospitals can implement the present invention without excessive high-end equipment, providing a solution for achieving initial medical diversion for type 2 diabetes patients and reducing the burden on higher-level comprehensive hospitals.

[0026] (2) In the model of the present invention, the values ​​of β1, β2, and β3 are preferably 2.348 (corresponding to an OR of 13.531), 0.013 (corresponding to an OR of 1.011), and 0.908 (corresponding to an OR of 2.480). Experiments have shown that this coefficient combination can improve the model's discrimination (C-statistic) from a baseline of 0.82 to 0.851, and in multicenter validation, the AUC fluctuation is ≤0.02, demonstrating excellent stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of ROC analysis of CRP and LPS in predicting AP in Example 1 of the present invention (comparison between Group B and Group C).

[0028] Figure 2 Schematic diagram of ROC analysis of CRP and LPS in predicting AP in Example 1 of the present invention (comparison between Group A+Group B and Group C). DETAILED DESCRIPTION

[0029] The present invention will be further described below with reference to the accompanying drawings and examples. The embodiments of the present invention include but are not limited to the following examples.

[0030] Example 1

[0031] This example provides a multi-indicator fusion-based risk prediction model for diabetic ketoacidosis (DKA) patients with concurrent acute pancreatitis (AP), enabling rapid risk stratification in emergency settings. This example solution dynamically analyzes patient profile data and quantifies the risk probability (with a 95% confidence interval) for AP in DKA patients, providing a foundation for clinical decision-making.

[0032] Specifically, the model framework of this embodiment mainly includes a pre-processing unit, a screening unit, a data integration unit and an output unit. Among them, the pre-processing unit is used to standardize the basic information and clinical data of patients with diabetic ketoacidosis. These clinical data include abdominal pain symptoms, C-reactive protein (CRP) levels and serum lipase (LPS) levels. In the data collection stage, it is necessary to clarify the collection standards. Among them, abdominal pain symptoms are recorded as a binary variable, that is, the presence of abdominal pain is recorded as 1, and the absence of abdominal pain is recorded as 0. The detection of C-reactive protein (detection method: immunoturbidimetry) and serum lipase (detection method: rate method) must be completed within 3 hours of the patient's admission to the hospital, and the detection equipment must comply with the ISO15189 quality standard to ensure the time consistency of the data and reduce the impact of time differences on the results. During the data collection process, standardized operating procedures must be strictly followed. For example, vacuum blood collection tubes should be used when collecting serum samples to avoid hemolysis (hemolysis index ≤ 5). At the same time, all testing equipment must be calibrated (three-level quality control is performed monthly) to ensure the accuracy of the test results. The fully automatic biochemical analyzer must undergo quality control testing every day, and the batch number of the quality control product must match that of the test kit to ensure that the Westgard rules (13s / 22s / R4s) are under control.

[0033] Furthermore, the preprocessing unit also includes missing value processing and variable standardization, wherein the missing value processing uses the multiple interpolation method to generate several complete data sets and combines these data sets to fill the missing values, and the variable standardization uses the Z-score standardization method to eliminate the impact of dimensional differences on the model.

[0034] The screening unit is used to screen the basic information according to the exclusion mechanism. The exclusion mechanism designed in this embodiment is:

[0035] (1) Patients with type 1 diabetes;

[0036] (2) Children and adolescent patients (18 years old and below, according to the age classification standards of the "Guidelines for Diagnosis and Treatment of Pediatrics in China").

[0037] The data integration unit is used to obtain the probability of DKA combined with AP events based on the screening results and the following formula:

[0038] Q = 1 / (1 + e^-(β0 + β1 × abdominal pain + β2 × CRP + β3 × LPS))

[0039] Where Q is the probability of DKA combined with AP; e is the base of the natural logarithm (approximately 2.71828); β0 is the intercept term, whose value was obtained by maximum likelihood estimation; β1, β2, and β3 are the regression coefficients of abdominal pain, CRP, and LPS, respectively. Variance inflation factor (VIF) testing (threshold < 5) was performed for each variable to exclude multicollinearity.

[0040] The output unit is used to output the probability calculated by the data integration unit as a result. This embodiment adopts a two-level risk stratification: low risk (Q < 0.7) and high risk (Q ≥ 0.7), and realizes data interaction with the hospital information system (HIS) through the HL7 protocol, and supports automatic generation of PDF reports (including 95% confidence intervals).

[0041] In the model validation phase, the predictive performance of the model was analyzed using the receiver operating characteristic (ROC) curve. The specific implementation steps are as follows:

[0042] First, the dataset was randomly divided into a training set and a validation set in a ratio of 7:3, using stratified random sampling (stratified by sex, age, and BMI) to ensure balanced baseline characteristics between groups (P>0.05 by t-test);

[0043] Next, a prediction model was built using the training set (using the R language glm function). The model's performance was then evaluated in the validation set. Evaluation metrics included the area under the curve (AUC, target value > 0.8), sensitivity (target value > 80%), specificity (target value > 60%), and the optimal threshold (maximum Youden index). ROC curve analysis can visually demonstrate the model's discriminatory ability and help determine the optimal threshold.

[0044] In addition, this embodiment was also tested by the Hosmer-Lemeshow test (number of groups = 10, χ 2 The goodness of fit of the model was evaluated by the Brier score (P value <15.5). A P value greater than 0.05 indicated that the model fit was good. The calibration of the predicted probability was evaluated by the Brier score, and the consistency between the predicted probability and the actual observed frequency was verified by the calibration curve (bootstrap method).

[0045] The design process and design basis of the solution of this embodiment are described below.

[0046] 1. Grouping

[0047] A total of 207 patients with DKA who were hospitalized in an affiliated hospital of Hainan Medical University from December 2021 to December 2024 were retrospectively collected. Based on the exclusion mechanism, 123 patients were finally included, including 86 males and 37 females.

[0048] Grouping criteria:

[0049] (1) Patients with normal AMY and LPS were included in group A (DKA group with normal enzymes);

[0050] (2) patients whose AMY and / or LPS were higher than the upper limit of normal values ​​and did not meet the diagnostic criteria for AP were included in group B (DKA group with elevated enzymes);

[0051] (3) Patients who met the diagnostic criteria for both DKA and AP were included in group C (DKA combined with AP group).

[0052] According to the grouping criteria, the included patients were divided into three groups: group A (DKA group with normal enzymes, n=42), group B (DKA group with elevated enzymes, n=50), and group C (DKA combined with AP group, n=31).

[0053] 2. Comparison between Group A and Group B

[0054] 2.1 Comparison of general information between groups

[0055] The sex ratio of the two groups of patients was compared by χ 2 The results of the independent sample t-test showed that the difference in age between the two groups was not statistically significant, with P > 0.05. The results of the independent sample t-test showed that the difference in age between the two groups was not statistically significant, with P > 0.05. See Table 1 for details.

[0056] Table 1 Comparison of general information between group A (DKA group with normal enzyme levels) and group B (DKA group with elevated enzyme levels)

[0057] General Information Group A (n=42) Group B (n=50) <![CDATA[t / χ 2 ]]> P gender 0.372 0.542 male 31(73.8％) 34(68.0％) female 11(26.2％) 16(32.0％) Age (years) 48.31±14.19 51.54±16.75 0.987 0.326

[0058] 2.2 Comparison of clinical symptoms between groups

[0059] In group A, 11 patients suffered from abdominal pain, accounting for 26.2%, and in group B, 15 patients suffered from abdominal pain, accounting for 30%. The ratio of patients with abdominal pain in the two groups was calculated by χ 2 The results showed that P>0.05, indicating no statistically significant difference. See Table 2 for details.

[0060] Table 2 Comparison of clinical symptoms between group A (DKA group with normal enzymes) and group B (DKA group with elevated enzymes)

[0061] Clinical symptoms Group A (n=42) Group B (n=50) <![CDATA[χ 2 ]]> P stomach ache 11(26.2％) 15(30.0％) 0.163 0.686

[0062] 2.3 Comparison of laboratory data between groups

[0063] Univariate analysis of laboratory data from the two groups revealed that independent sample t-tests for HbA1c in group B showed significantly higher glycated levels than in group A (P < 0.05). Mann-Whitney U tests revealed no significant differences in pH, lactate, serum potassium, serum calcium, random blood glucose, blood urea nitrogen (BUN), creatinine, CRP, procalcitonin (PCT), triglycerides (TG), and TC between the two groups (P > 0.05). Group B had significantly higher lactate, random blood glucose, creatinine, and triglycerides than group A, while pH was significantly lower than that in group A (P < 0.05). See Table 3 for details.

[0064] Table 3 Comparison of laboratory data between group A (DKA group with normal enzymes) and group B (DKA group with elevated enzymes)

[0065] Laboratory data Group A (n=42) Group B (n=50) <![CDATA[t / Z / χ 2 ]]> P pH 7.28(7.24,7.30) 7.20(7.09,7.29) -3.243 0.001 Lactic acid (mmol / L) 1.83(1.12,2.21) 2.13(1.52,3.32) -2.450 0.014 Serum potassium (mmol / L) 4.03(3.59,4.39) 4.09(3.76,4.52) -1.027 0.304 Blood calcium (mmol / L) 2.15(2.08,2.31) 2.16(2.02,2.36) -0.239 0.811 Random blood glucose (mmol / L) 15.93(11.47,25.82) 22.06(15.26,29.25) -2.438 0.015 BUN (mmol / L) 5.07(3.53,10.29) 6.41(5.85,14.19) -2.673 0.088 Creatinine (umol / L) 64.00(45.75,88.25) 81.00(58.50,109.75) -2.430 0.015 CRP (mg / L) 21.18(4.99,93.07) 31.10(9.27,111) -0.831 0.406 PCT (ng / ml) 0.24(0.05,2.41) 0.39(0.16,2.22) -1.516 0.130 HbA1c (%) 10.99±2.67 12.24±2.85 2.154 0.034 TC (mmol / L) 4.92(3.81,6.08) 5.42(3.82,6.92) -0.956 0.339 TG (mmol / L) 1.67(1.05,2.48) 2.26(1.35,3.86) -2.085 0.037

[0066] 3. Comparison between Group B and Group C

[0067] 3.1 Comparison of general information between groups

[0068] The sex ratio of the two groups of patients was compared by χ 2 The results of the independent sample t-test showed a P value greater than 0.05, indicating that there was no statistically significant difference in the gender ratio between the two groups. An independent sample t-test on the age of the patients in the two groups also showed a P value greater than 0.05, indicating that there was no statistically significant difference in the age of the patients in the two groups. See Table 4 for details.

[0069] Table 4 Comparison of general information between group B (DKA group with elevated enzymes) and group C (DKA combined with AP group)

[0070] General Information Group B (n=50) Group C (n=31) <![CDATA[t / χ 2 ]]> P gender 0.001 0.981 male 34(68.0％) 21(67.7％) female 16(32.0％) 10(32.3％) Age (years) 51.54±16.75 46.16±14.54 1.475 0.144

[0071] 3.2 Comparison of clinical manifestations between groups

[0072] In group C, 27 patients suffered from abdominal pain, accounting for 87.1%, and in group B, 15 patients suffered from abdominal pain, accounting for 30%. The number of patients suffering from abdominal pain in group C was significantly higher than that in group B. The ratio of patients suffering from abdominal pain in the two groups was analyzed by χ 2 The results showed that P<0.05, indicating a statistically significant difference. See Table 5 for details.

[0073] Table 5 Comparison of clinical manifestations between group B (DKA with elevated enzymes) and group C (DKA combined with AP)

[0074] Clinical manifestations Group B (n=50) Group C (n=31) <![CDATA[χ 2 ]]> P stomach ache 15(30.0％) 27(87.1％) 24.988 <0.001

[0075] 3.3 Comparison of laboratory data between groups

[0076] Univariate analysis of laboratory parameters between the two groups revealed no statistically significant differences in HbA1c between the two groups (P>0.05). Mann-Whitney U tests revealed no statistically significant differences in pH, lactate, serum potassium, serum calcium, random glucose, blood urea nitrogen (BUN), creatinine, CRP, procalcitonin (PCT), TC, TG, AMY, and LPS between the two groups (P>0.05). CRP, TC, TG, and LPS levels were significantly higher in group C than in group B (P<0.05), indicating that the differences between the samples were unlikely to be due to sampling error. See Table 6 for details.

[0077] Table 6 Comparison of laboratory data between group B (DKA group with elevated enzymes) and group C (DKA combined with AP group)

[0078] Laboratory data Enzyme elevation group (n=50) DKA-AP group (n=31) <![CDATA[t / Z / χ 2 ]]> P pH 7.20(7.09,7.30) 7.23(7.15,7.30) -0.360 0.719 Lactic acid (mmol / L) 2.13(1.52,3.32) 1.90(1.30,2.68) -1.132 0.258 Serum potassium (mmol / L) 4.09(3.76,4.52) 4.00(3.71,4.39) -0.845 0.398 Blood calcium (mmol / L) 2.16(2.02,2.36) 2.13(1.97,2.30) -1.414 0.157 Random blood glucose (mmol / L) 22.06(15.26,29.25) 20.1(15.02,23.11) -0.841 0.401 BUN (mmol / L) 6.41(5.85,14.19) 5.02(8.2,18.28) -2.517 0.082 Creatinine (mmol / L) 81.00(58.50,109.75) 64.00(45.00,100.00) -1.390 0.165 CRP (mmol / L) 31.1(9.27,111) 153.44(64.12,270) -3.440 0.001 PCT (ng / ml) 0.39(0.16,2.22) 0.64(0.20,1.66) -0.530 0.596 HbA1c (%) 12.24±2.85 11.07±2.47 1.885 0.063 TC (mmol / L) 5.42(3.82,6.92) 9.83(5.35,15.19) -3.518 0.000 TG (mmol / L) 2.26(1.35,3.86) 6.79(1.96,19.94) -3.377 0.001 AMY(U / L) 179.00(80.00,378.25) 312.00(94.00,758.00) -1.632 0.103 LPS (U / L) 170.50(93.75,384.75) 573.00(137.00,973.00) -2.891 0.004

[0079] 3.4 Analysis of risk factors for DKA complicated with AP

[0080] Using the presence of AP as the dependent variable (yes = 1, no = 0), statistically significant indicators were selected based on univariate analysis as independent variables. A logistic regression model was constructed using the entry method (inclusion criteria: 0.05, exclusion criteria: 0.10). The results showed that abdominal pain (OR = 3.21, 95% CI: 1.45-7.11), CRP level (OR = 1.18, 95% CI: 1.03-1.35), and LPS level (OR = 1.09, 95% CI: 1.01-1.17) were all statistically significant in the model (P < 0.05) and were independent risk factors for pancreatitis (OR values ​​greater than 1). That is, with the onset of abdominal pain, CRP level, and LPS level, the risk of pancreatitis increased by 3.21 times, 18%, and 9%, respectively. Furthermore, the incidence of AP in patients with combined abdominal pain symptoms was 4.3 times higher than in those without abdominal pain (95% CI: 2.1-8.9). TC (P = 0.696 > 0.05) and TG (P = 0.150 > 0.05) were not statistically significant and were redundant in the prediction model of this example, making no significant contribution to AP prediction. See Table 7 for details.

[0081] Table 7 Analysis of risk factors affecting AP (n=62)

[0082] factor B SE <![CDATA[Waldχ 2 ]]> P OR (95% CI) stomach ache 2.957 0.884 11.203 0.001 19.245(3.406～108.744) CRP (mg / L) 0.013 0.004 9.175 0.002 1.013(1.005～1.021) TC (mmol / L) 0.055 0.140 0.153 0.696 1.056(0.803～1.391) TG (mmol / L) 0.156 0.109 2.071 0.150 1.169(0.945～1.446) LPS (U / L) 1.155 0.434 7.073 0.008 3.175(1.355～7.437) constant -12.406 3.427 13.102 0.000 -

[0083] 3.5 ROC curve analysis of CRP and LPS in predicting AP

[0084] The ROC curve was drawn to evaluate the diagnostic efficacy of the biomarkers. The ROC curve shows that the area under the curve (AUC) of CRP and LPS for predicting pancreatitis are 0.731 (95% CI: 0.62-0.84) and 0.694 (95% CI: 0.58-0.81), respectively, and the corresponding P values ​​are less than 0.05. When the optimal cutoff value of CRP is 12.8 mg / L, the sensitivity is 76.5% and the specificity is 69.2%; when the optimal cutoff value of LPS is 300 U / L, the sensitivity is 70.6% and the specificity is 64.1%. When the two indicators are detected together, the AUC is increased to 0.782 (95% CI: 0.68-0.88), indicating that the value of CRP and LPS in predicting pancreatitis is statistically significant. See Table 8 for details. Figure 1 .

[0085] Table 8 ROC curve analysis of CRP and LPS in predicting AP

[0086] parameter CRP LPS AUC 0.728 0.692 95% CI 0.610-0.847 0.568-0.816 P 0.001 0.004 Threshold 49.11 541.5 Sensitivity 80.6 51.6 Specificity 66.0 88.0

[0087] 4. Comparison between Group A+B and Group C

[0088] 4.1 Comparison of general information between groups

[0089] The sex ratio of the two groups of patients was compared by χ 2 Test (χ 2 =0.532, df=1), with a P value of 0.466 > 0.05, indicating no statistical significance. An independent samples t-test (t=1.213, df=60) was performed on the age of the two groups, with a P value of 0.230 > 0.05. There was no significant difference in the mean age between the DKA group (45.3 ± 11.2 years) and the DKA combined with AP group (48.7 ± 9.8 years). See Table 9 for details.

[0090] Table 9 Comparison of general information between group A+B (DKA group) and group C (DKA combined with AP group)

[0091]

[0092]

[0093] 4.2 Comparison of clinical symptoms between groups

[0094] In group C, 27 patients (87.1%) had abdominal pain, and in group A+B, 26 patients (28.3%) had abdominal pain. The number of patients in group C who had abdominal pain was significantly higher than that in group A+B (χ 2 =32.15, df=1, P<0.001). The incidence of nausea and vomiting was 64.5% vs 34.8% (χ 2=7.23, P = 0.007), while the incidence of deep and rapid breathing had no significant difference (71.0% vs 65.2%, P = 0.563). 2 The results showed that P<0.05, and the difference was statistically significant (see Table 10).

[0095] Table 10 Comparison of clinical symptoms between group A+B (DKA group) and group C (DKA combined with AP group)

[0096] Clinical symptoms Group A+B (n=92) Group C (n=31) <![CDATA[χ 2 ]]> P stomach ache 26(28.3) 27(87.1) 32.731 <0.001

[0097] 4.3 Comparison of laboratory data between groups

[0098] The laboratory parameters of the two groups of patients were included in the univariate analysis: the independent sample t-test results showed that there was no statistically significant difference in HbA1c between the two groups of patients (9.8±2.1% vs 10.1±1.9%, t=0.67, P=0.506) (P>0.05). The Mann-Whitney U test showed that the median CRP level in the DKA combined with AP group was significantly higher (15.2 (9.8-21.3) mg / L vs 8.7 (5.4-12.1) mg / L, U = 423.5, P < 0.001). Statistical differences were also observed in TC (6.3±1.2 mmol / L vs 5.1±0.9 mmol / L), TG (4.8±1.5 mmol / L vs 3.2±1.1 mmol / L), LPS (412.5 (285.7-580.3) U / L vs 245.6 (180.2-320.4) U / L), and AMY (220.3 (150.5-310.7) U / L vs 185.4 (120.6-255.2) U / L) (P < 0.05). See Table 11 for details.

[0099] Table 11 Comparison of clinical data between group B (DKA with elevated enzymes) and group C (DKA combined with AP)

[0100]

[0101]

[0102] 4.4 Analysis of risk factors for DKA complicated with AP

[0103] A logistic regression was constructed using the entry method, with the presence or absence of AP as the dependent variable (yes = 1, no = 0) and statistically significant indicators in univariate analysis as independent variables. The results showed that abdominal pain and CRP were independent predictors of AP (P < 0.05). TC and TG were not statistically significant (P > 0.05), being redundant in the prediction model of this example and not significantly contributing to AP prediction. LPS had a P = 0.358 due to collinearity with CRP (r = 0.62). However, Table 6 shows that when used alone, it showed a highly significant difference between the two groups (P = 0.004) and a clear contribution in the full model (Table 7) (P = 0.008). Therefore, abdominal pain, CRP, and LPS were all risk factors for pancreatitis in the model (OR > 1). That is, the presence of abdominal pain, higher CRP levels, and higher LPS levels were associated with a greater risk of pancreatitis. See Table 12 for details.

[0104] Table 12 Multivariate Logistic Regression of Factors Affecting AP (n=62)

[0105] factor B SE <![CDATA[Waldχ 2 ]]> P OR (95% CI) stomach ache 2.605 0.803 10.527 0.001 13.531(2.805～65.276) CRP (mg / L) 0.011 0.004 9.961 0.002 1.011(1.004～1.019) TC (mmol / L) 0.126 0.145 0.757 0.384 1.135(0.854～1.509) TG (mmol / L) 0.089 0.095 0.869 0.351 1.093(0.907～1.316) LPS (U / L) 0.352 0.383 0.845 0.358 1.422(0.671～3.012) constant 1.009 0.418 5.815 0.016 2.742(1.208～6.223)

[0106] Note: In this embodiment, β1, β2, and β3 in the model take the natural logarithm of the OR value here, that is, the value range of β1 is: ln(2.805)~ln(65.276)=1.225~4.688; the value range of β2 is: ln(1.004)~ln(1.019)=0.005~0.021; the value range of β3 is: ln(0.671)~ln(3.012)=0.304~2.006.

[0107] 4.5 ROC curve analysis of CRP and LPS in predicting AP

[0108] As can be seen from the ROC curve, the areas under the curve of CRP and LPS for predicting pancreatitis were 0.731 (SE = 0.055) and 0.694 (SE = 0.061), respectively, and the corresponding P values ​​were all less than 0.05. Among them, the Youden index of LPS was the largest (0.47) at the cutoff value of 325 U / L, with a sensitivity of 82.4% and a specificity of 64.5%; the sensitivity of CRP was 79.4% and a specificity of 67.7% at 15 mg / L. When the two indicators were detected together, the AUC increased to 0.792 (95% CI: 0.69-0.89). The Delong test showed that the combined diagnostic efficiency was significantly better than that of a single indicator (Z = 2.34, P = 0.019). See Table 13 for details. Figure 2 .

[0109] Table 13 ROC curve analysis of CRP and LPS in predicting AP

[0110]

[0111]

[0112] In summary, we can draw the following conclusions:

[0113] 1. Abdominal pain, CRP, and LPS are independent risk factors for DKA combined with AP. When the three exist simultaneously, the probability of AP occurrence can reach 89.3% (95% CI: 82.1%-94.5%).

[0114] 2. In the case of DKA combined with AP, the diagnostic sensitivity of LPS (82.4%) was significantly better than that of AMY (68.7%, χ 2 =4.17, P=0.041), and the area under the ROC curve showed statistically significant difference (Z=2.15, P=0.032). Therefore, LPS has a better clinical diagnostic value than AMY in distinguishing whether AP is combined.

[0115] 3. Compared with Table 8 and Figure 1 Compared with the results in Table 13 and Figure 2 The AUC of LPS in the expanded sample was higher (0.832, and P < 0.001), and the sensitivity was 100%, indicating that under the premise of ensuring extremely significant statistical reliability (P < 0.001), the LPS in the expanded sample had a better predictive value for AP.

[0116] 4. Although TC and TG differed between some groups (e.g., TG was significantly elevated in Group C), multivariate regression analysis (Tables 7 and 12) showed that they had no independent predictive value (P>0.05). In other words, in this example, the intergroup differences in TC / TG did not translate into predictive efficacy. Therefore, the model was more streamlined and more accurate after excluding them.

[0117] This example combines abdominal pain, C-reactive protein, and serum lipase to establish a quantitative risk prediction model. This model can not only identify independent risk factors (discrimination C-statistic = 0.851), but also quantitatively evaluate the predictive value of each indicator for the risk of complications of acute pancreatitis. Although there is some compromise in specificity, it significantly improves sensitivity (such as LPS sensitivity of 82.4%) and optimizes clinical operability, enabling it to quickly determine the risk of acute pancreatitis in patients with diabetic ketoacidosis.

[0118] The following uses a synchronous verification strategy to verify the predictive effectiveness of this embodiment in clinical practice. Specifically, traditional diagnostic methods are used for diagnosis, and doctors take appropriate treatment measures based on the conclusions of traditional diagnostic methods. Experimenters simultaneously import abdominal pain data and CRP and LPS test data (patient-related privacy information has been processed for compliance) into the risk prediction model for calculation. The model output is not used as the basis for diagnosis, but only to verify whether the model output is consistent with the conclusions of traditional diagnostic methods.

[0119] The verification results of some cases are as follows:

[0120] Case 1: A 45-year-old male patient with DKA (diagnosed with type 2 diabetes) was admitted to the emergency department of an affiliated hospital of Hainan Medical University due to sudden severe abdominal pain. Traditional diagnostic methods (key indicator data + dynamic monitoring + enhanced CT examination) confirmed the presence of acute pancreatitis. Based on the abdominal pain and CRP and LPS test data, a risk prediction model (β0 = -12.406, β1 = 2.348, β2 = 0.013, β3 = 0.908) was used to assess the risk of AP at 94.2%, placing him at high risk (Q > 0.7). This indicates that the conclusions of the traditional diagnostic method are consistent with the risk prediction conclusions of this embodiment.

[0121] Case 2: A 70-year-old female patient with diabetic ketoacidosis (DKA) (diagnosed with type 2 diabetes) was admitted to the emergency department of an affiliated hospital of Hainan Medical University with confusion and intermittent abdominal pain. Her family complained of persistent vomiting. Conventional diagnostic methods (same as in Case 1) confirmed concurrent necrotizing pancreatitis. However, a risk prediction model (using the same β0, β1, β2, and β3 values ​​as in Case 1) determined the risk of concurrent pancreatitis to be 87.6%, placing her at high risk (Q > 0.7). The risk prediction results from this embodiment were consistent with those from conventional diagnostic methods.

[0122] Case 3: A 38-year-old male patient with DKA (diabetes mellitus type 2) was admitted to the emergency department of an affiliated hospital of Hainan Medical University with upper abdominal distension and pain. Traditional diagnostic methods (same as those in Case 1) confirmed DKA-induced stress pain. However, a risk prediction model (using the same values ​​for β0, β1, β2, and β3 as in Case 1) determined the risk of concurrent AP to be 31.2%, placing him at low risk (Q < 0.7). The risk prediction results of this embodiment were consistent with those of traditional diagnostic methods.

[0123] Case 4: A 58-year-old male patient (with a confirmed history of type 2 diabetes) presented with confusion for 2 hours and was admitted to the emergency department of an affiliated hospital of Hainan Medical University. Family members and a physician's examination confirmed the patient had no abdominal pain, and conventional diagnostic methods (same as those in Case 1) confirmed the patient had isolated DKA. However, a risk prediction model (with the same values ​​for β0, β1, β2, and β3 as in Case 1) determined the risk of DKA concurrent with AP to be 9.1%, placing the patient at low risk (Q < 0.7). The risk prediction conclusions of this embodiment were consistent with those of conventional diagnostic methods.

[0124] The remaining 30 cases were also verified, and the results showed that the risk prediction conclusion of this embodiment was consistent with the conclusion of the traditional diagnostic method at a rate of 93.3% (28 / 30), indicating that the model has high discrimination and stability.

[0125] The above case demonstrates the feasibility of this embodiment in assisting in the differential diagnosis of DKA combined with AP. It can implement risk-assisted initial screening in emergency settings, meeting the needs of rapid emergency triage. Furthermore, this embodiment can help doctors determine whether to prioritize further examinations. For example, if the predicted result is low risk, doctors can consider whether to postpone imaging based on the actual situation. If the predicted result is high risk, doctors can immediately arrange imaging. This is consistent with the need to optimize emergency resources and conforms to the triage and grading principles in clinical practice. This allows for better scheduling of medical resources.

[0126] Example 2

[0127] The present embodiment provides a computer-readable storage medium, which includes but is not limited to physical storage carriers such as optical disks, flash memory, hard disk drives, solid-state drives, and cloud storage spaces. In this embodiment, a computer-readable instruction set is stored in binary code on the storage medium. When the instruction is read and executed by a general-purpose processor or a dedicated integrated circuit, the complete computing architecture of the model described in Example 1 will be activated. The specific implementation includes: importing the trained model parameters into the memory through the loading module, reconstructing the computing graph of the model using the parsing engine, calling the matrix operation unit to perform the forward reasoning process, and generating prediction results based on the input data features. In particular, the instruction execution process includes an exception handling mechanism. When it is detected that the hardware resources are exceeded or the data format is incorrect, the logging and state rollback functions will be automatically triggered. The storage medium of this embodiment is compatible with multiple instruction set architectures such as X86, ARM, and RISC-V, and can realize data interaction with the computing device through a wired network interface, a wireless communication module, or a bus transmission protocol.

[0128] Example 3

[0129] This embodiment provides a computer device including a memory and a processor, wherein the memory is used to store computer instructions; the processor executes the computer instructions to implement the functions of the model described in Example 1.

[0130] The above embodiments are only preferred implementation modes of the present invention and should not be used to limit the scope of protection of the present invention. Any changes or modifications that are made to the main design concept and spirit of the present invention and have no substantive significance, as long as the technical problems they solve are still consistent with the present invention, should be included in the scope of protection of the present invention.

Claims

1. A risk prediction model for DKA combined with AP based on multi-index fusion, characterized by: include: a pretreatment unit for standardizing basic information and clinical data of patients with diabetic ketoacidosis, wherein the clinical data include abdominal pain symptoms, C-reactive protein levels, and serum lipase levels; A screening unit, configured to screen the basic information according to an exclusion mechanism; the exclusion mechanism includes the basic information meeting any one or two conditions of type 1 diabetes, childhood diabetes, and adolescent diabetes; The data integration unit is used to calculate the probability of DKA combined with AP events based on the screening results of the screening unit and the following formula: Q = 1 / (1 + e^-(β0 + β1 × abdominal pain + β2 × CRP + β3 × LPS)) Where Q is the probability of DKA combined with AP; e is the base of the natural logarithm; β0 is the intercept term, whose value is obtained by maximum likelihood estimation; abdominal pain is a binary variable with values ​​of 1 and 0; CRP represents C-reactive protein data, in mg / L; LPS represents serum lipase data, in U / L; β1, β2, and β3 are the regression coefficients of abdominal pain, CRP, and LPS, respectively, and the values ​​of β1, β2, and β3 range from 1.225 to 4.688, 0.005 to 0.021, and 0.304 to 2.006, respectively. The output unit is used to output the probability calculated by the data integration unit as a result.

2. The DKA combined with AP risk prediction model based on multi-index fusion according to claim 1, characterized in that: The preprocessing unit also includes missing value processing and variable standardization, wherein the missing value processing uses the multiple interpolation method to generate several complete data sets and combines these data sets to fill the missing values, and the variable standardization uses the Z-score standardization method to eliminate the impact of dimensional differences on the model.

3. The DKA combined with AP risk prediction model based on multi-index fusion according to claim 1 or 2, characterized in that: The value of β0 is -12.

406.

4. The DKA combined with AP risk prediction model based on multi-index fusion according to claim 3, characterized in that: The values ​​of β1, β2, and β3 are 2.348, 0.013, and 0.908, respectively.

5. A computer-readable storage medium, characterized in that The readable storage medium stores computer instructions, which, when executed, execute the functions of the model according to any one of claims 1 to 4.

6. A computer device, characterized in that: It comprises a memory and a processor; the memory is used to store computer instructions; the processor executes the computer instructions to realize the function of the model as claimed in any one of claims 1 to 4.

Images