Method and system for fever of unknown infection grading prognosis evaluation based on inflammation factor spectrum

Abstract

This invention discloses a method and system for grading the severity and assessing the prognosis of febrile unexamined infections based on inflammatory cytokine profiles, belonging to the field of medical information technology. This invention collects multi-dimensional inflammatory cytokine profiles and clinical baseline data from patients with febrile unexamined infections. After feature engineering preprocessing, an assessment model is trained using machine learning combined with an imbalanced learning framework. After validation, the system is deployed and can automatically output infection severity grading, prognostic risk scores, and the contribution of key factors. This invention overcomes the shortcomings of traditional single inflammatory indicators, such as insufficient specificity, low assessment accuracy, and weak prognostic predictive ability. It achieves comprehensive, accurate, and automated assessment of febrile unexamined infections, which can assist in early clinical intervention, optimize resource allocation, and improve the standardization of diagnosis and treatment.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and in particular to a method and system for graded prognostic assessment of fever-related infections based on inflammatory cytokine profiles. Background Technology

[0002] Fever of unknown origin is a common syndrome in clinical practice with complex etiologies and high diagnostic difficulty. It is clearly defined as a clinical condition characterized by fever lasting more than 3 weeks, oral temperature repeatedly exceeding 38.3°C, and an inability to establish a definitive diagnosis despite at least one week of detailed examination. Infection is one of the most common causes of fever of unknown origin. Rapidly and accurately assessing the severity of infection and predicting patient prognosis is crucial for guiding clinical treatment decisions, optimizing the allocation of medical resources, and improving patient care outcomes.

[0003] Current clinical assessments of the severity of infectious fever of unknown origin primarily rely on vital signs, organ dysfunction scores such as the Sequential Organ Failure Assessment (SOFA), and a few laboratory inflammatory markers such as C-reactive protein (CRP) and procalcitonin (PCT). These traditional assessment methods have several inherent limitations: First, single inflammatory markers are easily affected by non-infectious factors, lack detection specificity, and cannot accurately reflect the true infection status; Second, the existing assessment system has limited ability to identify early infection, atypical infection and mixed infection, making it difficult to accurately classify the severity of infection. Third, there is a lack of automated tools in clinical practice that can integrate multi-dimensional inflammatory information and quantify prognostic indicators such as complication risk, hospital stay, and mortality. Prognostic assessment relies on clinical experience, which is highly subjective and has limited accuracy.

[0004] In summary, traditional assessment methods are insufficient to meet the precise diagnosis and treatment needs of patients with fever and unknown infections. Clinically, there is an urgent need for a new technical solution that can comprehensively, objectively, and automatically complete the grading of infection severity and prognosis assessment. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for grading and prognostic assessment of fever-related infections based on an inflammatory cytokine spectrum. By systematically detecting and analyzing a set of inflammatory factors closely related to infection and immune response, a machine learning model is constructed to achieve automated and quantitative grading of infection severity and accurate assessment of patient prognosis.

[0006] To achieve the above objectives, this invention provides a method for graded prognostic assessment of fever-related infections based on inflammatory cytokine profiles, comprising the following steps: S1. Collect demographic information, vital signs, history of underlying diseases and blood samples of patients with fever of unknown origin upon admission. Detect blood samples to obtain multidimensional inflammatory factor spectrum data. At the same time, collect infection severity grading and prognostic outcomes as label data. S2. Encode and standardize the raw data to construct training and test sets; S3. Select a machine learning algorithm, combine it with an imbalanced learning framework to train and optimize it, and obtain an infection grading and prognosis assessment model. S4. Evaluate the model performance on the test set and compare it with traditional evaluation methods for verification; S5. Deploy the validated model, input patient data, and automatically output infection severity grading, prognostic risk score, and contribution of key inflammatory factors.

[0007] Preferably, the multidimensional inflammatory cytokine spectrum includes at least interleukin-6 (IL-6), tumor necrosis factor-α (TNF-α), interferon-γ (IFN-γ), C-reactive protein (CRP), and procalcitonin (PCT).

[0008] Preferably, the machine learning algorithm in step S3 includes support vector machine, logistic regression, random forest, and XGBoost; the imbalanced learning framework uses random downsampling of majority class samples and ensemble base models to handle the data class imbalance problem.

[0009] Preferably, in step S4, AUC, accuracy, sensitivity, and specificity are used to evaluate the model performance, and the evaluation differences between this model and traditional methods are compared using the Bootstrap sampling method.

[0010] A system for grading and prognostic assessment of fever-related infections based on inflammatory cytokine profiles includes: The data acquisition and access module is used to connect to the hospital's laboratory information system and electronic medical record system to automatically or semi-automatically collect patients' inflammatory factor data and clinical information. The data preprocessing and feature engineering module is used to complete data cleaning, encoding, and standardization, and to build usable feature vectors for the model. The core analysis engine module has a built-in pre-trained infection grading and prognosis assessment model. It receives processed feature data and performs infection grading and prognosis risk assessment calculations. The results visualization and report generation module is used to display the evaluation results and generate structured evaluation reports; The model update and maintenance module is used to perform incremental training and optimization of the model based on newly labeled data.

[0011] Preferably, the core analysis engine module outputs results including infection severity grading, prognostic risk score, and ranking of the contribution of key inflammatory factors.

[0012] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: 1. More comprehensive and accurate assessment: By integrating a multi-dimensional spectrum of inflammatory factors, it overcomes the shortcomings of insufficient specificity of single indicators, and can more sensitively reflect the infection status and the intensity of the immune response, thus achieving a more objective severity classification.

[0013] 2. Enhanced prognostic prediction capabilities: Based on machine learning models, it can uncover the complex nonlinear relationship between inflammatory factors and long-term prognosis, identify high-risk patients in advance, and assist clinicians in early intervention and resource allocation.

[0014] 3. Automation and standardization: The entire assessment process can be completed automatically by the system, reducing subjective differences from humans, improving assessment efficiency and consistency, and contributing to the standardization of clinical diagnosis and treatment.

[0015] 4. Interpretability: The system can output the importance ranking of each inflammatory factor, helping doctors understand the basis of the model's decision-making, increasing clinical trust, and potentially revealing new biomarkers.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the prognostic assessment method for fever-related infections based on inflammatory cytokine spectrum, as described in Embodiment 1 of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Example 1 like Figure 1As shown, the prognostic assessment method for fever-related infections of unknown origin based on inflammatory cytokine profiles includes the following steps: S1: Data Acquisition and Preprocessing Clinical data were collected from patients with fever of unknown origin upon admission, including demographic information (age, sex), vital signs, history of underlying diseases, and blood samples. Blood samples were tested to obtain multidimensional inflammatory cytokine profile data, including interleukin-6 (IL-6), tumor necrosis factor-α (TNF-α), interferon-γ (IFN-γ), C-reactive protein (CRP), and procalcitonin (PCT). Simultaneously, the final infection diagnosis and severity classification (e.g., classified as mild, moderate, or severe based on the Sequential Organ Failure Assessment (SOFA) score) confirmed by etiology or comprehensive clinical diagnosis, as well as the patient's prognostic outcome at discharge (e.g., cured, improved, not cured, or deceased), were collected as label data for model training.

[0022] S2: Feature Engineering and Dataset Construction The collected raw data is preprocessed. Categorical variables (such as gender and underlying medical history) are converted into numerical variables using one-hot encoding. Continuous inflammatory factor values ​​are standardized or normalized to eliminate the influence of dimensions. The processed data is randomly divided into a training set and an independent test set according to a preset ratio (e.g., 7:3). The training set is used for model building and parameter tuning, while the test set is used for final model performance validation.

[0023] S3: Machine Learning Model Training and Optimization Multiple machine learning algorithms were selected as candidate models, such as Support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF), and Gradient Boosting Decision Trees (e.g., XGBoost). To address the common outcome class imbalance problem in clinical data (e.g., the number of patients with good prognosis far exceeds the number of those with poor prognosis), an ensemble imbalanced learning framework was employed. For SVM, LR, and RF, samples from the majority class were randomly downsampled to generate multiple balanced subsets, which were then used to train base models before ensemble integration. K-fold cross-validation (e.g., five-fold cross-validation) was performed on the training set and repeated multiple times to evaluate model stability and optimize hyperparameters. The model with the best average performance on the validation set was selected as the final grading and prognostic assessment model.

[0024] S4: Model Validation and Performance Evaluation The performance of the final model was evaluated on an independent test set. Evaluation metrics included the area under the receiver operating characteristic (AUC), accuracy, sensitivity, and specificity. To further validate the superiority of the model over traditional scoring methods (such as those based solely on SOFA scores or the PCT single metric), a bootstrap sampling method was used for repeated sampling (e.g., 1000 times), and a 95% confidence interval was calculated for the difference in AUC between the new model and the traditional method. If this confidence interval did not contain 0, the new model was considered to significantly outperform the traditional method.

[0025] S5: Application of Tiered Assessment and Prognostic Prediction Deploy the validated model as an assessment system. For new patients with fever of unknown origin, input their inflammatory cytokine profile and baseline clinical data upon admission, and the system will automatically output: 1) infection severity grading results (e.g., low risk, intermediate risk, high risk); 2) prognostic risk scores (e.g., probability of complication, risk of in-hospital mortality); 3) key influencing factor analysis, showing the contribution of each inflammatory factor to the current assessment results, providing a reference for clinical intervention.

[0026] Example 2 This embodiment provides a fever-based prognostic assessment system for suspected infection classification based on inflammatory cytokine spectrum, used to implement the method in Embodiment 1, comprising: Data acquisition and access module: It interfaces with the hospital laboratory information system (LIS) and electronic medical record system (EMR) to automatically or semi-automatically collect patient inflammatory factor detection data and clinical information.

[0027] Data preprocessing and feature engineering module: Cleans, encodes, and standardizes the raw data to construct feature vectors usable by the model.

[0028] Core analysis engine module: It has a built-in pre-trained machine learning model that is responsible for receiving processed feature data and performing severity classification and prognostic risk assessment calculations.

[0029] Results visualization and report generation module: Visualizes the model output's grading results, risk scores, and key factors in the form of charts, dashboards, etc., and generates structured assessment reports.

[0030] Model update and maintenance module: Provides an interface that allows incremental training and performance optimization of the model after obtaining new batches of labeled data, ensuring the timeliness and accuracy of the model.

[0031] Example 3 This embodiment uses three years of clinical data from a top-tier hospital to validate the model and obtain quantitative data for practical application, as detailed below: 1. Clinical data, inclusion and exclusion criteria, and a complete set of indicators and methods for detecting inflammatory factors. This application strictly followed the clinical definition of fever of unknown origin to select study subjects, including adult patients with fever lasting more than three weeks, oral temperature exceeding 38.3°C multiple times and no clear cause after a week of systematic examination, finally diagnosed with an infectious cause, and who could complete data collection and inflammatory factor testing within 24 hours of admission. At the same time, cases with malignant tumors, autoimmune diseases, long-term use of immunosuppressants, use of antibiotics or glucocorticoids within the past 72 hours, pregnancy or lactation, and severe liver or kidney failure, which may significantly interfere with the expression of inflammatory factors and prognosis, were excluded. All study subjects signed informed consent forms.

[0032] Five hundred patients with fever of unknown origin who were eventually diagnosed with infectious diseases were included in this study, including 278 males and 222 females, aged 18-85 years, with a mean age of 56.3 years. Clinical data collected included baseline information such as age, sex, history of underlying diseases, vital signs, complete blood count, liver and kidney function, and coagulation function. The highest SOFA score within 28 days of admission was used as the severity grading standard (≤2 points for mild, 3-5 points for moderate, and ≥6 points for severe). The 28-day clinical outcome (cured / improved for good, not cured / death for poor) was used as the prognostic label. Inflammatory factor detection employed a complete set of indicators consisting of IL1β, IL6, IL8, IL10, TNFα, IFNγ, CRP, PCT, and SAA. Serum was separated by centrifugation through intravenous blood collection upon admission, and multifactorial simultaneous quantitative detection was achieved using liquid chromatography-suspended chip technology. Standards and quality control materials were used throughout the process to ensure that the intra-batch coefficient of variation was less than 10% and the inter-batch coefficient of variation was less than 15%, ensuring standardized and reliable test results.

[0033] 2. Machine learning model training parameters, comparative experimental data, and statistical test results. This application included 500 patients with infectious fever of unknown origin who met the criteria. They were randomly divided into a training set of 400 cases and an independent test set of 100 cases at a ratio of 7:3. Five-fold cross-validation was used to optimize the model parameters. To address the class imbalance problem caused by the small number of poor prognostic samples in the clinical data, a balanced dataset was constructed by combining SMOTE oversampling and majority class downsampling. An ensemble learning model was built based on logistic regression, support vector machine, random forest, and XGBoost. Categorical variables were one-hot encoded and continuous variables were Z-score standardized.

[0034] The final model achieved a macro-average AUC of 0.91 and an overall accuracy of 85% in the three-class classification task of infection severity. In the binary classification prediction of poor prognosis, the AUC was 0.88, with a sensitivity of 86% and a specificity of 89%. Feature importance ranking showed that IL6, PCT, and IL10 were the core predictors, and all performance indicators were significantly better than traditional methods such as SOFA scoring alone, CRP combined with PCT, and single machine learning models. Bootstrap 1000 repeated sampling statistical test showed that the AUC difference between this model and traditional assessment methods did not include 0 in the 95% confidence interval, indicating a statistically significant difference.

[0035] 3. Evaluate the system's architecture, interfaces, deployment, and application details. The evaluation system implemented in this application adopts a modular microservice architecture, consisting of five main modules: data acquisition and access, data preprocessing and feature engineering, core analysis engine, result visualization and report generation, and model update and maintenance. It seamlessly integrates with hospital electronic medical record systems and laboratory information systems via RESTful APIs and HL7 standard interfaces. It can automatically or semi-automatically collect patient inflammatory factors and clinical baseline data, supporting real-time assessment of single cases and offline processing of batch data. The system possesses automated preprocessing capabilities such as data cleaning, missing value imputation, standardization, and feature screening. The core engine has a built-in integrated evaluation model that can quickly complete infection grading and prognostic risk calculations. Results are displayed in graphical visualization and support the export of structured PDF reports. The system supports local deployment on hospital intranet servers and private cloud platforms, and is equipped with data anonymization, hierarchical access control, and operation log auditing functions. Furthermore, it can perform incremental model training and iterative optimization based on newly labeled clinical data, ensuring long-term application stability and accuracy.

[0036] 4. Efficacy data and validation results in actual clinical application In external validation with 100 independent test cases, the system achieved an overall accuracy of 85% in grading infection severity, 88% sensitivity in identifying severe infections, and 86% sensitivity in predicting adverse prognostic events at 28 days. Its overall assessment efficacy is significantly superior to traditional single inflammatory markers and conventional clinical scoring tools. In practical clinical applications, the system can complete fully automated assessment output within 10 seconds of patient data upload, significantly reducing manual assessment time and improving consistency among different physicians. It can identify high-risk, severe infection patients early, providing objective evidence for timely initiation of monitoring, optimization of antibiotic use, and rational allocation of critical care resources. This effectively reduces the rate of missed diagnoses and delayed intervention in high-risk cases, demonstrating stable and reliable application value in the standardized diagnosis and treatment of patients with fever and unexplained infections, early risk stratification, and accurate prognostic assessment.

[0037] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for graded prognostic assessment of fever-related infections based on inflammatory cytokine profiles, characterized in that, The steps are as follows: S1. Collect demographic information, vital signs, history of underlying diseases and blood samples of patients with fever of unknown origin upon admission. Detect blood samples to obtain multidimensional inflammatory factor spectrum data. At the same time, collect infection severity grading and prognostic outcomes as label data. S2. Encode and standardize the raw data to construct training and test sets; S3. Select a machine learning algorithm, combine it with an imbalanced learning framework to train and optimize it, and obtain an infection grading and prognosis assessment model. S4. Evaluate the model performance on the test set and compare it with traditional evaluation methods for verification; S5. Deploy the validated model, input patient data, and automatically output infection severity grading, prognostic risk score, and contribution of key inflammatory factors.

2. The method for evaluating the prognosis of a fever of unknown origin due to an infectious disease based on the classification of an inflammatory factor profile according to claim 1, characterized by: The multidimensional inflammatory cytokine spectrum includes at least interleukin-6 (IL-6), tumor necrosis factor-α (TNF-α), interferon-γ (IFN-γ), C-reactive protein (CRP), and procalcitonin (PCT).

3. The method for evaluating the prognosis of a fever of unknown origin infection in stages based on an inflammation factor spectrum according to claim 1, characterized in that: The machine learning algorithms mentioned in step S3 include support vector machine, logistic regression, random forest, and XGBoost; the imbalanced learning framework uses random downsampling of majority class samples and ensemble base models to handle the data class imbalance problem.

4. The method for evaluating the prognosis of a fever of unknown origin infection in stages based on an inflammation factor spectrum according to claim 1, characterized in that: In step S4, AUC, accuracy, sensitivity, and specificity are used to evaluate the model performance, and the evaluation differences between this model and traditional methods are compared using the Bootstrap sampling method.

5. A system for implementing a method for the graded prognostic evaluation of febrile infections of unknown etiology based on the profile of inflammatory factors as claimed in any one of claims 1-4, characterized in that, include: The data acquisition and access module is used to connect to the hospital's laboratory information system and electronic medical record system to automatically or semi-automatically collect patients' inflammatory factor data and clinical information. The data preprocessing and feature engineering module is used to complete data cleaning, encoding, and standardization, and to build usable feature vectors for the model. The core analysis engine module has a built-in pre-trained infection grading and prognosis assessment model. It receives processed feature data and performs infection grading and prognosis risk assessment calculations. The results visualization and report generation module is used to display the evaluation results and generate structured evaluation reports; The model update and maintenance module is used to perform incremental training and optimization of the model based on newly labeled data.

6. The inflammation factor profile-based fever of unknown origin infection staging prognosis evaluation system according to claim 5, characterized in that: The core analysis engine module outputs results including infection severity grading, prognostic risk scores, and ranking of the contribution of key inflammatory factors.

Images