Method for establishing mortality risk prediction index for acute diquat pesticide poisoning

Abstract

The present invention belongs to the field of prognosis of pesticide poisoning. Specifically disclosed is a method for establishing a mortality risk prediction index for acute diquat pesticide poisoning, comprising: collecting data; establishing a mortality risk index model based on demographic characteristics and routine blood test results; and training, internally validating, and externally validating the model. The mortality risk index provided by the present invention comprehensively considers age, blood concentration of diquat, white blood cell count, and aspartate aminotransferase concentration, aiming to accurately evaluate, in the form of an index, the mortality risk of patients suffering from acute diquat pesticide poisoning. Through in-depth analysis of large-scale clinical data, the mortality risk index of the present invention possesses scientific validity and accuracy at the data level, and can also play an important role in clinical practice. Further, the present invention can be applied to construct a terminal device for predicting the mortality risk of acute diquat pesticide poisoning, assisting medical teams in more reasonably allocating medical resources and prioritizing high-risk patients, thereby improving overall treatment outcomes and the patients' chances of survival.

Description

A method for establishing a mortality risk index for acute diquat pesticide poisoning Technical Field

[0001] This invention belongs to the field of pesticide poisoning prognosis, and specifically discloses a method for establishing a predictive mortality risk index for acute diquat pesticide poisoning. Background Technology

[0002] Poisoning is a global public health problem, and effectively controlling poisoning incidents is crucial for reducing morbidity and mortality. The American Association of Poison Control Centers points out that the United States reports more than 2 million poisoning cases annually, with a total morbidity rate as high as 633 per 100,000 people. Data shows that from 2009 to 2013, poisoning was consistently the fifth leading cause of death in China, primarily related to suicide attempts rather than accidental poisoning.

[0003] Paraquat and its newly emerging alternative, diquat, are toxic to organs such as the kidneys and liver. Studies have found that the mortality rate of acute diquat poisoning is approximately 18.60%–60.00%, with oral ingestion being the primary route of poisoning. Due to the lack of an effective antidote, emergency department treatment for diquat poisoning primarily involves gastric lavage, adsorption, catharsis, and total bowel lavage; however, these measures may be difficult to implement in underdeveloped areas. The World Health Organization's International Chemical Safety Program is the only guideline providing a lethal dose of diquat, defining it as 6–12 grams without considering other factors such as exposure time and age. Admittedly, due to the lack of prognostic assessment for diquat poisoning patients, clinical decision-making often struggles to implement individualized treatment plans. Summary of the Invention

[0004] To address the aforementioned issues, this invention discloses a predictive mortality risk index for acute diquat pesticide poisoning and its establishment method. This index provides scientific support for clinical decision-making based on demographic characteristics and routine blood test results. The index established based on this method can accurately predict the mortality risk of patients with acute diquat poisoning.

[0005] This invention includes the following technical solutions:

[0006] A method for establishing a mortality risk index for acute diquat pesticide poisoning includes the following steps:

[0007] Step 1: Collect data and build a queue

[0008] Data including patients’ demographic characteristics and routine blood test results were collected to construct a cohort of patients with acute diquat pesticide poisoning.

[0009] Step two: Establish a mortality risk index based on demographic characteristics and routine blood test results.

[0010] The collected demographic characteristics and blood routine test results were used as candidate predictive variables. The importance of the features was evaluated by three algorithms: Boruta algorithm, univariate Cox model, and stepwise Cox regression. Variables that could demonstrate predictive value in all three algorithms were included in the mortality risk index. A mortality risk index model based on demographic characteristics and blood routine test results was established.

[0011] Step 3: Model training, internal validation, and external validation

[0012] The queue data established in step one is used as the development set. The data is divided into training set and test set at a certain ratio to train and internally validate the mortality risk index model, and external validation is performed on the queue data other than the development set.

[0013] Furthermore, in the above-mentioned method for establishing a predictive mortality risk index for acute diquat pesticide poisoning, step two includes:

[0014] The Boruta algorithm generates a random forest iteratively and uses the important features determined in the previous round and their corresponding shadow features to calculate the importance of new features, thereby determining the important features;

[0015] The univariate Cox model uses a p-value < 0.05 as the threshold for selecting variables;

[0016] The combination of variables that achieves the minimum AIC is obtained by stepwise Cox regression.

[0017] Furthermore, in the aforementioned method for establishing a mortality risk index for acute diquat pesticide poisoning, the demographic characteristic is age.

[0018] Furthermore, in the aforementioned method for establishing a predictive mortality risk index for acute diquat pesticide poisoning, the blood routine test results include the initial diquat blood concentration, white blood cell count, and aspartate aminotransferase concentration.

[0019] Furthermore, in the above-mentioned method for establishing a mortality risk index for acute diquat pesticide poisoning, in step three, the ratio of training set to test set data is 7:3.

[0020] Furthermore, in the above-mentioned method for establishing a mortality risk index for acute diquat pesticide poisoning, step two involves incorporating age, initial diquat blood concentration, white blood cell count, and aspartate aminotransferase (AST) concentration into the model to obtain a mortality risk index based on demographic characteristics and routine blood test results, as shown below:

[0021] Death Risk Index=100-(98.2031318439685)*(3.8342×Age+0.0136601*Concentration+6.7254×WBC+0.2535×AST)

[0022] Where Age: age (years); Concentration: initial diquat blood concentration (ng / mL); WBC: white blood cell count (*10^6). 9 / L); AST: Aspartate aminotransferase concentration (U / L); * indicates exponentiation, × indicates multiplication.

[0023] Furthermore, in the above-mentioned method for establishing a mortality risk index for acute diquat pesticide poisoning, the higher the mortality risk index, the more severe the patient's condition and the greater the probability of death; the cutoff value for the survival outcome prediction based on the mortality risk index is 81, that is, if the calculated mortality risk index is ≤81, the predicted outcome for the patient is survival, and conversely, if the calculated mortality risk index is >81, the predicted outcome for the patient is death.

[0024] The present invention also discloses a mortality risk index for acute diquat pesticide poisoning, which is established by the method described in any of the above-mentioned methods.

[0025] This invention also discloses a device for predicting mortality risk in acute diquat pesticide poisoning, comprising:

[0026] Data filtering module: used to collect data including patients' demographic characteristics and blood routine test results, and to construct personal data of patients with acute diquat pesticide poisoning;

[0027] Scoring module: The patient's individual data is scored using the acute diquat pesticide poisoning prediction mortality risk index mentioned above;

[0028] Probability prediction module: Compares the scores in the scoring module with the cutoff values ​​for survival outcome predictions in the database, and outputs the prediction results.

[0029] The present invention also discloses a terminal device for predicting mortality risk in acute diquat pesticide poisoning, characterized in that it includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it calls the aforementioned acute diquat pesticide poisoning mortality risk index.

[0030] The present invention has the following beneficial effects:

[0031] Based on demographic characteristics and routine blood test results, this invention proposes an index model for predicting the risk of death in patients with acute diquat pesticide poisoning. This model comprehensively considers multiple key factors, including the patient's age, diquat blood concentration, white blood cell count, and aspartate aminotransferase (AST) level, thus providing accurate prognostic predictions for poisoned patients and ensuring the reliability and accuracy of the assessment. Through in-depth analysis of large-scale clinical data, the mortality risk index of this invention not only possesses scientific validity and accuracy at the data level but also plays a crucial role in clinical practice. Specifically, this index provides healthcare professionals with an important decision-making tool during treatment, enabling them to develop more personalized and effective treatment plans based on the patient's specific condition, particularly demographic characteristics and routine blood test results. Simultaneously, in emergency medical situations, this index can help medical teams allocate medical resources more rationally, prioritizing high-risk patients, thereby improving overall treatment outcomes and patient survival rates. Furthermore, the mortality risk index of this invention is highly adaptable and practical. It can not only be widely applied in different types of medical institutions but can also be adjusted and optimized based on continuously updated clinical data to ensure its accuracy and effectiveness in various clinical scenarios. This index allows clinicians to better understand the pathological process of diquat poisoning, optimize treatment strategies, and ultimately improve patients' prognosis and quality of life. Attached Figure Description

[0032] Figure 1 shows the ROC curves in an embodiment of the present invention;

[0033] Figure 2 shows the confusion matrix in an embodiment of the present invention;

[0034] Figure 3 is a line diagram in an embodiment of the present invention. Detailed Implementation

[0035] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Unless otherwise specified, the reagents or instruments used in the embodiments of this invention are all commercially available conventional reagent products.

[0037] The data used in this embodiment comes from the Jiangsu Provincial Medical Association's Poisoning Branch cohort, including patient demographics and blood routine test results obtained from the hospital's electronic medical record system, as well as 28-day prospective follow-up outcomes. This project uses data from teaching hospitals of Nanjing Medical University for model development (training and internal validation), and uses patient data from other hospitals for external validation.

[0038] This study (Ethics No.: 2021-SR-394, ClinicalTrial.gov ID: NCT 05215457, October 7, 2017) was approved by the Medical Ethics Committee of the First Affiliated Hospital of Nanjing Medical University and conducted in accordance with the principles of the Declaration of Helsinki. This study involved the analysis of routinely collected data; therefore, a waiver of written informed consent was obtained from the institutional review committee. To protect patient privacy, all information was anonymized prior to analysis.

[0039] Inclusion criteria were as follows: (1) a history of oral exposure to diquat reported by the patient or their legal representative; (2) a toxicology test completed immediately upon admission; (3) complete blood routine test results; and (4) the patient or their legal representative (in the case of the patient being unconscious) was aware of and agreed to the treatment plan. Exclusion criteria were as follows: (1) the patient had ingested other toxins besides diquat; (2) no diquat pesticide residues were detected in blood or urine samples, and no clinical symptoms were observed; and (3) a history of major primary heart, lung, liver, kidney, or brain diseases.

[0040] Ultimately, 104 patients from teaching hospitals of Nanjing Medical University were included in the development set, and data from 98 patients from the Jiangsu Provincial Medical Association Poisoning Branch cohort (excluding those in the development set) were included in the external validation set.

[0041] Example

[0042] A method for establishing a mortality risk index for acute diquat pesticide poisoning includes the following steps:

[0043] Step 1: Collect data and build a queue

[0044] Data collected included demographic characteristics and complete blood count (CBC) results, and a cohort of patients with acute diquat pesticide poisoning was constructed. Demographic characteristics included age and sex; blood test results included initial diquat blood concentration, aspartate aminotransferase (AST), platelet count, hemoglobin count, red blood cell count, white blood cell count, blood urea nitrogen, neutrophil count, monocyte count, serum creatinine concentration, lymphocyte count, basophil count, and eosinophil count.

[0045] Step two: Establish a mortality risk index based on demographic characteristics and routine blood test results.

[0046] The collected demographic characteristics and routine blood test results were used as candidate predictive variables, and the importance of the features was evaluated using three algorithms: Boruta algorithm, univariate Cox model, and stepwise Cox regression.

[0047] First, the Boruta algorithm is used to filter these candidate variables. Boruta is a feature selection algorithm based on random forests. Its key feature is that it generates multiple "shadow variables," which are meaningless data generated by randomizing the original data. The random forest model is trained with all candidate variables and these shadow variables. The algorithm determines whether a variable is an important feature by comparing the importance scores of each actual variable with those of its shadow variables. Specifically, the Boruta algorithm iteratively builds multiple random forest models, gradually eliminating actual variables whose scores are lower than those of their shadow variables. Finally, variables whose scores are significantly higher than those of their shadow variables are retained; these variables are considered to contribute significantly to the survival outcome. The advantage of this step is that it can capture the non-linear relationships between complex data from multiple dimensions, filtering out more comprehensive candidate variables.

[0048] Secondly, univariate Cox regression analysis is performed to assess the association between variables and survival outcomes. In this step, each variable is analyzed individually using Cox regression to evaluate the strength of its association with the prognostic event. By calculating the hazard ratio (HR) and its corresponding p-value for each variable, it is possible to determine which variables are statistically significant in the univariate analysis. A p-value less than 0.05 is set as the screening threshold, meaning only those variables that are statistically significant and associated with prognosis are retained. This step helps to further narrow down the range of candidate variables, removing those variables that, when viewed individually, do not have a significant impact on prognosis.

[0049] Finally, stepwise Cox regression analysis was performed. This step uses stepwise regression to optimize the variable combination based on the AIC (Akaike Information Criterion). AIC is a standard for evaluating model quality, seeking a balance between model fit and complexity. Stepwise Cox regression repeatedly adds and removes variables, continuously comparing the AIC value of each model to find the model with the smallest AIC. The final variable combination not only has strong explanatory power but also avoids overfitting, ensuring the model's generalization ability. During stepwise regression, the introduction or removal of variables is dynamic; the algorithm continuously adjusts until the model's AIC value reaches its optimal state. This step helps to build a concise and accurate model. Among the three algorithms, age, diquat blood concentration, white blood cell count, and aspartate aminotransferase level all demonstrated predictive value and were included in the mortality risk index.

[0050] The following is the predicted mortality risk index for acute diquat pesticide poisoning:

[0051] Death Risk Index=100-(98.2031318439685)*(3.8342×Age+0.0136601*Concentration+6.7254×WBC+0.2535×AST)

[0052] Where Age: age (years); Concentration: initial diquat blood concentration (ng / mL); WBC: white blood cell count (*10^6). 9 / L); AST: Aspartate aminotransferase concentration (U / L); * indicates exponentiation, × indicates multiplication.

[0053] Step 3: Model training, internal validation, and external validation

[0054] A Cox regression model was used to train the development set, which was then divided into training and internal validation datasets in a 7:3 ratio to construct a mortality risk index for acute diquat pesticide poisoning. The model fit was evaluated using the C-index and area under the curve (AUC), with 0.5 indicating random prediction and 1.0 indicating perfect accuracy. Receiver operating characteristic (ROC) curves, positive predictive value (PPV), negative positive value (NPV), accuracy, and confusion matrix were used to further evaluate model performance. All statistical analyses were performed using R software (version 4.1.2). Statistical tests were two-tailed, with a significance level of α = 0.05.

[0055] As shown in Table 1, the development group included 104 patients with a mortality rate of 36.5%. 56% were male. Significant differences were found between survivors and deceased patients in the development group in exposure history (reported diquat exposure amount and duration), initial diquat blood concentration, white blood cell count, monocyte count, basophil count, red blood cell count, hemoglobin, alanine aminotransferase (ALT), aspartate aminotransferase (AST) concentrations, blood urea nitrogen (BUN), serum creatinine concentration, and neutrophil-to-lymphocyte ratio. All parameters were comparable between the training and test sets (Table S2). The external validation set included 98 patients, comprising 69 survivors and 29 deceased patients. The majority were female (61.2%). Survivors (median age: 20 years) were younger than deceased patients (median age: 30 years). Exposure history (self-reported diquat exposure amount and exposure time), initial diquat blood concentration, white blood cell count, monocyte count, red blood cell count, hemoglobin, alanine aminotransferase, aspartate aminotransferase concentration, and serum creatinine concentration differed significantly between surviving and deceased patients.

[0056] Table 1. Comparison of demographic characteristics, exposure history, and biomarkers of surviving and deceased patients with acute diquat poisoning.

[0057] The 28-day mortality risk increased with age (hazard ratio: 1.04 [1.02, 1.08]), with initial diquat blood concentration (hazard ratio: 1.14 [1.06, 1.23]), with white blood cell count (hazard ratio: 1.07 [1.03, 1.11]), and with aspartate aminotransferase (AST) level (hazard ratio: 1.002 [1.001, 1.004]). As shown in Figure 1, the concordance index for predicting mortality risk in acute diquat pesticide poisoning was 0.86±0.06 (training set), 0.92±0.07 (internal validation set), and 0.87±0.07 (external validation set); the area under the ROC curve was 0.92 [0.85, 0.98] (training set), 0.93 [0.84, 1] (internal validation set), and 0.90 [0.83, 0.97] (external validation set). As shown in Figure 2, the PPV for all three datasets was 1, indicating no false positives (predicting survival patients as having died). The NPVs for the training, testing, and external validation sets were 0.75, 0.77, and 0.77, respectively. Furthermore, the model achieved a training accuracy of 0.81, a testing accuracy of 0.78, and an external validation accuracy of 0.79.

[0058] Once trained, the mature model can be used to predict the mortality risk of patients with diquat poisoning. As shown in Figure 3, when using it, first, based on the patient's specific data (including age, initial diquat blood concentration, white blood cell count, and aspartate aminotransferase concentration), locate the corresponding numerical positions of each variable in the nomogram, and draw a vertical line downwards intersecting the "score" axis. After reading the score of each variable on the "score" axis, add the scores together to obtain the total score. Then, draw a vertical line downwards from the position corresponding to the total score on the "total score" axis to find the mortality probability on the "Acute Diquat Pesticide Poisoning Predictive Mortality Risk Index" axis. This mortality risk index uses 81 points as the dividing line for predicting mortality risk: when the total score is greater than or equal to 81 points, the patient is predicted to be at high risk, with a significantly increased probability of death; while when the total score is less than 81 points, the patient is predicted to be at low risk, with a lower probability of death. This risk value can help clinicians more accurately assess the patient's prognosis and provide guidance in the formulation of treatment plans.

[0059] The above embodiments illustrate and describe the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for establishing a predictive mortality risk index for acute diquat pesticide poisoning, characterized in that, Includes the following steps: Step 1: Collect data and build a queue Data including patients’ demographic characteristics and routine blood test results were collected to construct a cohort of patients with acute diquat pesticide poisoning. Step two: Establish a mortality risk index based on demographic characteristics and routine blood test results. The collected demographic characteristics and blood routine test results were used as candidate predictive variables. The importance of the features was evaluated by three algorithms: Boruta algorithm, univariate Cox model, and stepwise Cox regression. Variables that could demonstrate predictive value in all three algorithms were included in the mortality risk index. A mortality risk index model based on demographic characteristics and blood routine test results was established. Step 3: Model training, internal validation, and external validation The queue data established in step one is used as the development set. The data is divided into training set and test set at a certain ratio to train and internally validate the mortality risk index model, and external validation is performed on the queue data other than the development set.

2. The method for establishing a predictive mortality risk index for acute diquat pesticide poisoning according to claim 1, characterized in that, In step two: The Boruta algorithm generates a random forest iteratively and uses the important features determined in the previous round and their corresponding shadow features to calculate the importance of new features, thereby determining the important features; The univariate Cox model uses a p-value < 0.05 as the threshold for selecting variables; The combination of variables that achieves the minimum AIC is obtained by stepwise Cox regression.

3. The method for establishing a predictive mortality risk index for acute diquat pesticide poisoning according to claim 1, characterized in that, The demographic characteristic mentioned is age.

4. The method for establishing a predictive mortality risk index for acute diquat pesticide poisoning according to claim 1, characterized in that, The blood routine test results include the initial diquat blood concentration, white blood cell count, and aspartate aminotransferase concentration.

5. The method for establishing a predictive mortality risk index for acute diquat pesticide poisoning according to claim 1, characterized in that, In step three, the ratio of training set to test set data is 7:

3.

6. The method for establishing a predictive mortality risk index for acute diquat pesticide poisoning according to claim 1, characterized in that, In step two, age, initial diquat blood concentration, white blood cell count, and aspartate aminotransferase (AST) concentration are incorporated into the model to obtain a mortality risk index based on demographic characteristics and routine blood test results, as shown below: Death RiskIndex = 100 - (98.2031318439685) * (3.8342 × Age + 0.0136601 * Concentration + 6.7254 × WBC + 0.2535 × AST) Where Age: age (years); Concentration: initial diquat blood concentration (ng / mL); WBC: white blood cell count (*10^6). 9 / L); AST: Aspartate aminotransferase concentration (U / L); * indicates exponentiation, × indicates multiplication.

7. The method for establishing a predictive mortality risk index for acute diquat pesticide poisoning according to claim 6, characterized in that, The higher the mortality risk index, the more severe the patient's condition and the greater the probability of death. The cutoff value for the survival outcome prediction based on the mortality risk index is 81. That is, if the calculated mortality risk index is ≤81, the patient's predicted outcome is survival; conversely, if the calculated mortality risk index is >81, the patient's predicted outcome is death.

8. A predictive mortality risk index for acute diquat pesticide poisoning, characterized in that, It is established by the method described in any one of claims 1-7.

9. A device for predicting mortality risk in acute diquat pesticide poisoning, characterized in that, include: Data filtering module: used to collect data including patients' demographic characteristics and blood routine test results, and to construct personal data of patients with acute diquat pesticide poisoning; Scoring module: The acute diquat pesticide poisoning prediction mortality risk index as described in claim 8 is used to score individual patient data; Probability prediction module: Compares the scores in the scoring module with the cutoff values ​​for survival outcome predictions in the database, and outputs the prediction results.

10. A terminal device for predicting mortality risk in acute diquat pesticide poisoning, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, invokes a predictive mortality risk index for acute diquat pesticide poisoning as described in claim 8.

Images