An intelligent evaluation system for medical consumables fusing evidence-based medicine and personalized analysis

Abstract

The present application relates to the field of medical information technology, in particular to a medical consumable intelligent evaluation system fusing evidence-based medicine and personalized analysis. The system comprises a data fusion module, an evidence-based evidence management module, a personalized analysis module, a decision interaction module and a closed-loop feedback tracing module. The data fusion module integrates internal and external data of the hospital to generate a structured data set; the evidence-based evidence management module quantitatively grades the evidence and constructs a dynamic evidence database; the personalized analysis module combines patient characteristics and preferences, and outputs individualized multi-dimensional analysis results through a patient-surgery scene-consumable matching model; the decision interaction module generates structured options based on the analysis results, supports doctor-patient collaborative decision-making and records the process; the closed-loop feedback tracing module collects real-world feedback data for calibrating and iteratively optimizing the matching model. The present application realizes intelligent, reasonable and personalized evaluation of medical consumables, and improves the scientificity, interpretability and auditability of the evaluation.

Description

A smart evaluation system for medical consumables that integrates evidence-based medicine and personalized analysis Technical Field

[0001] This invention relates to the field of medical information technology, specifically to an intelligent evaluation system for medical consumables that integrates evidence-based medicine and personalized analysis. Background Technology

[0002] Medical consumables, such as cardiac stents, artificial joints, and surgical staplers, are widely used in clinical surgical procedures. Their selection directly impacts treatment outcomes, complication risks, surgical efficiency, and patient medical costs. With the rapid iteration of consumable products and intensified competition among similar products, multiple consumable options often exist for the same surgical scenario. Furthermore, different patients exhibit significant differences in underlying diseases, risk factors, anatomical variations, affordability, and personal preferences. Therefore, how to scientifically evaluate candidate consumables for specific patients and surgical scenarios and develop interpretable selection criteria has gradually become a crucial technical requirement for hospital clinical decision support, doctor-patient communication, and refined consumable management.

[0003] Current medical consumable evaluation technologies largely rely on clinical expert experience, manufacturer data, and fixed indicator tables. These are typically used for macro-level scenarios such as access, procurement, or usage management, but they struggle to reflect patient and surgical scenario differences, and evaluation criteria can easily vary depending on the personnel and institution. In recent years, clinical decision support systems based on natural language processing and machine learning have gradually developed. Clinicians can use these systems to analyze medical records more deeply, leading to the most appropriate treatment decisions. Clinicians can input information and wait for the system to output the "correct" decision, using simple outputs to guide the decision-making process. However, existing intelligent systems primarily focus on disease treatment plans or drug selection, lacking an engineered solution for individualized selection of medical consumables that systematically integrates real-world hospital data, external evidence-based data, patient characteristics, surgical scenario parameters, and patient preferences. Existing clinical decision support systems often struggle to structure evidence elements into computable data formats and convert evidence quality into weight parameters that can be used for model fusion, thus lacking a stable mapping mechanism from group evidence to individual predictions. Furthermore, the selection of consumables involves communication between doctors and patients and the boundaries of responsibility. Without solidified records and integrity verification of input data, evidence chains, model versions and intermediate outputs, the decision-making process is difficult to audit and trace.

[0004] Based on the current state and shortcomings of existing technologies, there is an urgent need for an intelligent evaluation technology solution for medical consumables that integrates evidence-based medicine and personalized analysis to solve the above-mentioned technical problems and improve the scientificity, interpretability and auditability of medical consumable selection. Summary of the Invention

[0005] This invention provides an intelligent evaluation system for medical consumables that integrates evidence-based medicine and personalized analysis, aiming to solve the technical problems mentioned in the background art.

[0006] This invention provides an intelligent evaluation system for medical consumables that integrates evidence-based medicine and personalized analysis, comprising:

[0007] The data fusion module is used to obtain patient diagnosis and treatment data, consumable information and corresponding evidence-based medicine evidence from internal and external data sources of the hospital, and generate structured datasets;

[0008] The evidence-based evidence management module is used to quantify and rate the evidence-based medical evidence in the structured dataset according to preset evidence quality assessment rules, and to build an evidence-based evidence library associated with consumable identification.

[0009] The personalized analysis module is used to target the surgical scenario corresponding to the patient's medical data, combine the patient's medical data with the patient's willingness to use consumables, retrieve matching evidence from the evidence-based evidence library, and output multi-dimensional analysis results for multiple optional consumables through a pre-established patient-surgical scenario-consumable matching model.

[0010] The decision interaction module is used to generate structured decision options based on the multi-dimensional analysis results, and supports medical staff and patients to jointly confirm the target consumables plan and record the confirmation results through a visual interactive interface;

[0011] The closed-loop feedback traceability module is used to collect real-world clinical feedback data under the application of consumables corresponding to the confirmation result, align the real-world clinical feedback data with the multi-dimensional analysis results to form calibration data, and iteratively optimize the patient-surgical scenario-consumable matching model based on the calibration data.

[0012] Furthermore, the data fusion module includes:

[0013] The multi-source data acquisition unit is used to acquire patient diagnosis and treatment data and consumable circulation data from hospital information systems and supply chain management systems through standard data interfaces, and to acquire evidence-based medicine texts related to the target consumables from external medical literature databases through network acquisition mechanisms.

[0014] The data standardization unit is used to map the names of diagnoses, surgeries and consumables to standard codes based on a medical terminology mapping table, and to perform named entity recognition and relation extraction on unstructured text to generate structured records.

[0015] The associated building unit is used to associate the same consumable with the unique identifier of the consumable as the primary key, and to build a patient-surgical-consumable associated data chain.

[0016] A converged storage unit is used to store the structured record and the associated data chain.

[0017] Furthermore, the structured records generated by the data standardization unit include in-hospital patient diagnosis and treatment data and evidence element data, wherein:

[0018] The patient's medical data includes individual information data, preoperative assessment data, surgical scenario parameters, perioperative medical data, postoperative follow-up data, and cost and resource usage data. Among them, individual information data includes age, gender, body mass index, and underlying diseases; preoperative assessment data includes the patient's imaging and physiological examination results, preoperative risk assessment conclusions, and willingness to use consumables; and surgical scenario parameters include the type, name, method, grade, and duration of the surgery.

[0019] The evidence element data is automatically extracted from evidence-based medicine evidence texts, including study design type, study subject characteristics, intervention measures, control measures, key outcome indicators and corresponding statistical data, and forms a standardized set of evidence elements containing RCT data and RWD data.

[0020] Furthermore, the evidence-based management module includes:

[0021] An automated quality rating unit is used to perform quantitative quality assessments on RCT data and RWD data respectively. Specifically: for RCT data, key outcome indicators are assessed based on machine-executable evidence rating rules in the dimensions of risk of bias, inconsistency, indirectness, imprecision, and publication bias to generate a preliminary quality rating; for RWD data, a quantitative scoring table is used to score data from the dimensions of data source, study design, bias control, and consistency, and the scores are normalized to obtain a credibility weight; and a final quality rating for consumable evidence is generated based on the preliminary quality rating and credibility weight.

[0022] The evidence base construction unit is used to encapsulate the final quality rating and corresponding evidence element data into standardized evidence items and store them in the evidence-based evidence base.

[0023] Furthermore, the evidence-based management module also includes:

[0024] The evidence dynamic monitoring and updating unit is used to periodically search external medical literature databases and obtain new evidence texts related to consumables in the evidence-based evidence library. After triggering the structured processing and final quality rating of the new evidence, it is written into the evidence-based evidence library.

[0025] The evidence version management and traceability unit is used to manage the versions of evidence items corresponding to the same consumable, retain historical versions and record evidence update logs to form a traceable evidence evolution chain.

[0026] Furthermore, the personalized analysis module includes:

[0027] The feature and intention processing unit extracts patient diagnosis and treatment data from the structured dataset and quantifies individual information, surgical scenario parameters, and patient consumable usage intention data from the patient diagnosis and treatment data into patient-surgical scenario feature vectors and preference weight vectors.

[0028] The multi-dimensional prediction unit is used to retrieve evidence element data that matches the candidate consumables from the evidence-based evidence library based on the patient-surgical scenario feature vector, and output analysis results including efficacy prediction, risk prediction and cost-benefit prediction through a pre-established patient-surgical scenario-consumable matching model.

[0029] The scoring synthesis unit is used to convert the analysis results of efficacy prediction, risk prediction and cost-benefit prediction into scores according to standardized rules, and to perform weighted fusion with the preference weight vector to generate multi-dimensional analysis results including clinical value score, economic value score and comprehensive recommendation index.

[0030] Furthermore, the patient-surgical scenario-consumable matching model includes a treatment efficacy prediction sub-model, a risk prediction sub-model, and a cost-benefit prediction sub-model; wherein,

[0031] The efficacy prediction sub-model is configured to: use the average efficacy data of the target consumable in the evidence-based evidence library and the feature vector of the study population as a benchmark, calculate the difference between the patient-surgical scenario feature vector and the feature vector of the study population, input the difference into the trained efficacy adjustment model to output the efficacy adjustment amount, and then combine the efficacy adjustment amount with the average efficacy data of the group to obtain the individualized efficacy prediction value of the patient.

[0032] The risk prediction sub-model is configured to: identify patient-specific risk enhancement factors based on patient-surgical scenario feature vectors, and input the baseline risk data of the target consumables in the evidence-based evidence library into the risk prediction model, and output the individualized conditional probability of the preset adverse events;

[0033] The cost-benefit prediction sub-model is configured to: calculate the expected total medical cost based on the individualized efficacy prediction value and the individualized conditional probability simulation of different clinical outcome paths, combine the target consumable procurement cost, related surgical and anesthesia costs and adverse event handling costs, and compare the expected total medical cost with the health output gain to output the incremental cost-benefit ratio.

[0034] Furthermore, the decision-making interaction module includes:

[0035] The decision option generation unit is used to encapsulate each candidate consumable into a structured decision option object based on the evidence-based evidence base and the multi-dimensional analysis results. The structured decision option object includes a comprehensive recommendation index, a summary of key advantages and disadvantages, and a final quality rating of the consumable evidence.

[0036] The interface rendering unit is used to render differentiated interactive interfaces according to the user's role. The physician interface is used to display a comparison matrix of candidate consumables and provide a detailed view linked to efficacy, risk, cost and evidence summary. The patient interface is used to present the differences in a non-professional way and in a visual manner and provide preference input controls.

[0037] The confirmation and recording unit is used to record the consumable plan jointly confirmed by the doctor and the patient, the confirmation time, the information of the participants, and the key decision-making considerations.

[0038] Furthermore, the closed-loop feedback traceability module includes:

[0039] The tracking identifier generation and feedback collection unit is used to generate a globally unique decision tracking identifier for each consumable confirmation, and to collect information on actual therapeutic effects, actual occurrence of adverse events, and actual costs from the hospital information system and follow-up system based on the tracking identifier.

[0040] The prediction-fact alignment unit is used to align the collected actual therapeutic effects, actual occurrence of adverse events, and actual cost information with the prediction results output by the personalized analysis module to form calibration data pairs.

[0041] The model iteration unit is used to trigger the retraining of the matching model and update the model parameters when the calibration data pairs reach a preset number or deviation threshold.

[0042] Furthermore, the closed-loop feedback traceability module also includes:

[0043] The report solidification unit is used to solidify and generate cryptographic hash values ​​based on the multi-dimensional analysis results, the patient-surgical scenario-consumable matching model version, the content recorded by the confirmation record unit, and the content generated and collected by the tracking identifier generation and feedback collection unit, based on the actual consumables selected by the specific patient. The report is then packaged and output in a preset audit format to support a full-chain traceable evaluation report for third-party integrity verification.

[0044] The present invention has the following beneficial effects:

[0045] 1. This invention achieves scientific rigor in the evaluation of medical consumables, improving the objectivity and comparability of evaluation results. By structuring and quantifying evidence-based medicine, this invention enables evidence from different sources and study designs to participate in model calculations with uniform weights, thereby enhancing the objectivity and comparability of evaluation results. Furthermore, by constructing a patient-surgical-consumable matching model that integrates evidence-based medicine and individual patient characteristics, the evaluation process is based on verifiable medical evidence and real clinical data, avoiding the subjective biases inherent in existing technologies that rely solely on experience or static indicators. By structuring and quantifying randomized controlled trial data and real-world data, evidence from different research sources and designs can participate in calculations in a unified manner, and its influence can be reasonably adjusted according to the quality of evidence, thus improving the reliability of evaluation results in terms of statistical significance and medical evidence, and enhancing the scientific rationality of the medical consumables evaluation method.

[0046] 2. The interpretability of medical consumable evaluation has been achieved, enabling personalized analysis of medical consumable use for specific patients. By incorporating evidence-based medicine, individual patient treatment data, surgical scenario parameters, and the economic attributes of consumables into a unified data processing and computation framework, the evaluation of consumables has shifted from group average conclusions to individualized patient decisions, overcoming the problem in existing technologies where evidence-based evidence cannot directly guide the selection of consumables for specific patients. By decomposing the consumable evaluation process into interconnected but functionally defined sub-models such as efficacy prediction, risk prediction, and cost-benefit prediction, the formation of evaluation results has a clear data source and logical path. Each prediction result can be traced back to the corresponding patient treatment characteristics, surgical scenario parameters, and evidence-based evidence items, avoiding the problem of simply outputting black-box recommendations without supporting evidence. By presenting the results of the analysis of different consumables in terms of efficacy, safety, and cost in a structured manner, and combining evidence quality information and patient preference weights to form a comprehensive recommendation index, physicians and patients can intuitively understand the differences between different consumable options and the reasons for these differences. This improves understanding and trust in the consumable selection process, supports collaborative decision-making between doctors and patients, and helps avoid medical disputes related to consumable selection.

[0047] 3. This invention achieves auditability and traceability in the evaluation of medical consumables. Through a closed-loop feedback traceability mechanism, it systematically records and manages the data, evidence, and models used in the evaluation and confirmation process of consumables, ensuring that each consumable selection has a complete chain of technical evidence. By associating and storing patient treatment data, evidence-based evidence items, evidence quality rating results, and model prediction outputs, and generating traceable records during decision confirmation, the evaluation process can be fully reconstructed and verified afterward. Furthermore, by aligning real-world feedback data generated during actual clinical applications with prediction results to form calibration data for model iterative optimization, this not only improves the long-term stability of the model but also provides an objective data foundation for supervision, quality management, and accountability, thereby enhancing the usability and credibility of the medical consumables evaluation system in medical management and auditing scenarios. Attached Figure Description

[0048] Figure 1 is a structural block diagram of a medical consumables intelligent evaluation system that integrates evidence-based medicine and personalized analysis provided by the present invention;

[0049] Figure 2. Schematic diagram of the consistency verification between the efficacy prediction results and real-world follow-up efficacy in the patient-surgical scenario-consumable matching model. Detailed Implementation

[0050] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0051] As shown in Figure 1, this embodiment provides an intelligent evaluation system for medical consumables that integrates evidence-based medicine and personalized analysis, including:

[0052] The data fusion module is used to obtain patient diagnosis and treatment data, consumable information and corresponding evidence-based medicine evidence from internal and external data sources of the hospital to generate structured datasets.

[0053] Specifically, the data fusion module includes: a multi-source data acquisition unit, used to acquire patient diagnosis and treatment data and consumable circulation data from hospital information systems and supply chain management systems through standard data interfaces, and to acquire evidence-based medical evidence text related to the target consumable from external medical literature databases through a network acquisition mechanism; a data standardization unit, used to map diagnosis, surgery, and consumable names to standard codes based on a medical terminology mapping table, and to perform named entity recognition and relation extraction on unstructured text to generate structured records; an association construction unit, used to associate the same consumable's user patients, surgical scenarios, cost data, and clinical feedback data with the consumable's unique identifier as the primary key, constructing a patient-surgical scenario-consumable association data chain; and a fusion storage unit, used to store the structured records and the association data chain. The patient diagnosis and treatment data includes individual patient information, preoperative assessment data, surgical scenario parameters, perioperative diagnosis and treatment data, postoperative follow-up data, and cost and resource usage data. The evidence element data is automatically extracted from evidence-based medicine evidence texts, including study design type, study subject characteristics, intervention measures, control measures, key outcome indicators and corresponding statistical data, and forms a standardized set of evidence elements containing RCT (randomized controlled trial) data and RWD (real world data) data.

[0054] More specifically, the data fusion module begins with a multi-source data acquisition unit. This unit interfaces with the hospital information system and supply chain management system via communication interfaces to acquire patient diagnosis and treatment data and consumable circulation data. Simultaneously, this unit periodically accesses external medical literature databases, such as PubMed or the Cochrane Library, through pre-defined web crawlers or application programming interfaces to retrieve evidence-based medical evidence texts related to the target consumables. This raw data constitutes the initial input to the system, and its inherent heterogeneity and unstructured nature present technical obstacles that need to be overcome in subsequent processing.

[0055] To address the issue of data heterogeneity, the data standardization unit transforms the collected raw data. This unit uses a pre-built medical terminology mapping table to uniformly map diagnostic names, surgical procedure names, and consumable product names from different systems to a standard medical coding system, such as ICD-10, CPT, or UDI. For unstructured text data, such as discharge summaries or medical literature abstracts, this unit deploys a trained named entity recognition model. This model can automatically identify and extract key entities from the text, such as "patient age 65 years," "using total hip arthroplasty," and "postoperative infection rate reduced by 5%," and uses relation extraction techniques to determine the semantic relationships between these entities; for example, "reduced" is associated with "total hip arthroplasty" and "postoperative infection rate." Through this process, unstructured text is transformed into structured records containing clearly defined fields and values. Specifically, data from within the hospital is structured into patient treatment data, which includes individual information extracted from the hospital information system, preoperative assessment data, surgical scenario parameters, perioperative treatment data, postoperative follow-up data, and cost and resource usage data. Individual information data includes age, gender, body mass index, and underlying diseases. Preoperative assessment data includes the patient's imaging and physiological examination results, preoperative risk assessment conclusions, and willingness to use surgical consumables. Surgical scenario parameters include the type, name, method, grade, and duration of the surgery. It should be noted that willingness to use surgical consumables includes the patient's cost sensitivity, risk tolerance, and recovery speed preference. Data from external literature is structured into evidence element data, which includes automatically extracted data on study design type, study subject characteristics, intervention measures, control measures, key outcome indicators, and corresponding statistical measures. Based on this, a standardized set of evidence elements is formed, including RCT data and RWD data.

[0056] To establish inherent connections between data, the association construction unit performs key data association operations. This unit uses the unique identifier of the consumable as the primary key, which can be the unique identifier of a medical device. Using this primary key as the core, the system automatically associates all relevant usage records of the consumable within the hospital, including the specific patient who used the consumable, the specific surgical scenario in which it was applied, the actual cost data, and subsequent clinical feedback data. This series of association operations constructs a complete "patient-surgical scenario-consumable association data chain." For example, for a specific batch of artificial knee joint prostheses, this data chain can trace back to the medical records of all patients who used the prosthesis, details of their total knee replacement surgery, hospitalization costs, and joint function scores six months post-surgery. Finally, the fusion storage unit persistently stores the standardized and associated structured records and association data chain in the database, forming a unified and well-organized structured dataset for subsequent modules to access.

[0057] In the data fusion module, the named entity recognition and relation extraction algorithm relied upon by the data standardization unit directly targets medical terminology and clinical expressions in its training data and model optimization objectives. Its function is to accurately parse clinical text to generate high-quality structured records, providing accurate input for subsequent rule-based quality assessment and vector-based similarity calculations. The association construction unit, using the unique identifier of the consumable as the primary key, is not simply a data linking mechanism, but rather reflects a deep understanding of the regulatory need for full lifecycle traceability of individual medical consumables. Its function is to establish a traceable data topology, providing an indispensable data foundation for prediction and fact-based alignment based on the same identifier in the closed-loop feedback traceability module. Through the synergistic effect of the above data collection, standardization, association, and storage, the data fusion module transforms data scattered across different systems and in different formats into a unified data foundation for consumable evaluation. This provides reliable data support for the subsequent establishment of a patient-surgical-consumable matching model and solves the technical problems of heterogeneous data sources, difficulty in reusing evidence, and the inability to systematically associate clinical feedback in existing technologies.

[0058] Furthermore, the system also includes an evidence-based management module, which is connected to the data fusion module and is used to quantify and rate the evidence-based medical evidence in the structured dataset according to preset evidence quality assessment rules, and to build an evidence-based evidence library associated with consumable identification.

[0059] Specifically, the evidence-based management module includes: an automated quality rating unit, used to perform quantitative quality assessments on RCT data and RWD data respectively, wherein: for RCT data, the key outcome indicators are assessed based on machine-executable evidence rating rules in the dimensions of risk of bias, inconsistency, indirectness, imprecision, and publication bias to generate a preliminary quality grade; for RWD data, a quantitative scoring table is used to score from the dimensions of data source, study design, bias control, and consistency, and the scores are normalized to obtain a credibility weight; and a final quality rating for consumable evidence is generated based on the preliminary quality grade and credibility weight; an evidence library construction unit, used to encapsulate the final quality rating and corresponding evidence element data into standardized evidence entries and store them in the evidence-based evidence library; an evidence dynamic monitoring and updating unit, used to periodically search external medical literature databases and obtain new evidence texts related to consumables in the evidence-based evidence library, triggering structured processing and final quality rating of the new evidence before writing it into the evidence-based evidence library; and an evidence version management and traceability unit, used to manage the versions of evidence entries corresponding to the same consumable, retain historical versions, and record evidence update logs to form a traceable evidence evolution chain.

[0060] More specifically, the evidence-based management module receives a structured dataset from the data fusion module, which already contains evidence element data extracted from external medical literature. The core function of this module is to objectively and reproducibly quantify the quality of this original evidence and construct a dynamic, traceable knowledge base. The automated quality rating unit employs differentiated evaluation strategies for two types of evidence data. For randomized controlled trial data, the automated quality rating unit processes it according to pre-defined, machine-executable rules that are transformed into conditional judgment logic. In the bias risk assessment dimension, the system parses the structured fields in the evidence element data: for "study design," if it is "randomized, double-blind, placebo-controlled," it is marked as a low-bias risk starting point; for "blinding implementation," the system determines the blinding status of subjects and researchers by matching keywords in the descriptive text; for "allocation concealment," it assigns logical values ​​based on descriptions such as "central randomization" and "sealed envelope"; for "loss to follow-up rate," it directly reads the numerical field and compares it with a pre-defined threshold; if the loss to follow-up rate exceeds 20%, a risk label is triggered; for "intention-to-treat analysis," it judges based on the presence of an "ITT analysis" description. The system assigns a score or risk label to each judgment point and performs cumulative or downgrade logical operations on the risk of bias dimension. In the inconsistency dimension, the system compares the statistical heterogeneity of outcomes from different subgroups within the same study, or compares the effect size differences between different studies. In the indirectness dimension, the system assesses the fit between PICO elements in the evidence and the current clinical problem, such as comparing the age range of the study population with the age of the target patients. In the imprecision dimension, the system reads the confidence intervals of the outcome indicators; if the interval is too wide or contains an invalid line, it is marked. In the publication bias dimension, information such as the existence of pre-registration and the source of research funding can be used for auxiliary judgment. After the assessment of each dimension is completed, the system synthesizes the results according to a set of predefined decision rules. These rules specify how the combination of risk labels from different dimensions leads to an increase or decrease in the overall grade, ultimately outputting a preliminary quality grade such as "high," "medium," "low," or a numerical grade corresponding to 1 to 5.

[0061] For real-world data, the automated quality rating unit uses a structured quantitative scoring table. In the data source dimension, a base score is assigned based on the data source type, such as "prospective registry database," "retrospective electronic medical record," or "medical insurance claims database" from external medical literature repositories. Additional points are added or deducted based on data completeness and validation mechanisms. In the research design rigor dimension, research types such as "cohort study," "case-control study," or "cross-sectional study" are graded and scored, and the adequacy of the sample size is assessed. In the confounding bias control method dimension, the use and proper description of statistical control methods such as "multivariate regression," "propensity score matching," or "instrumental variables" are evaluated, and scores are given based on the complexity and applicability of the methods. In the consistency with other studies dimension, the system compares the key effect size of the RWD study with the results of other existing studies on the same issue in the evidence base, calculating statistical measures of consistency or making qualitative judgments. After scoring each dimension, the system performs a weighted sum according to preset weight coefficients, and then maps the total score to a confidence weight between 0 and 1 using a normalization function. The closer the weight value is to 1, the higher the confidence of the RWD evidence.

[0062] After obtaining the preliminary quality rating of the RCT and the credibility weight of the RWD, this unit performs evidence integration. The system maintains an integration mapping rule; for example, it maps a "high" quality RCT to a baseline credibility of 0.9, and a "medium" quality RCT to 0.7. For RWD evidence, its credibility weight can be used directly. When multiple types of evidence exist for the same clinical question, the system can generate a final quality rating according to a preset strategy, such as adopting the highest level of evidence principle or weighted averaging of the credibility of multiple pieces of evidence, outputting a unified numerical rating or grade label that can be used for subsequent model calculations. This process transforms the subjective, qualitative evidence evaluation in the literature into objective, computable quality indicators within the system.

[0063] The evidence base construction unit performs an encapsulation operation based on the above rating results. It encapsulates the final quality rating along with the corresponding complete evidence element data, such as the characteristics of the research subjects, intervention measures, and statistical measures of outcome indicators, into a standardized evidence item object. This object, with the unique identifier of the consumable as the core index key, is stored in the evidence-based evidence base, thereby transforming discrete literature data into structured knowledge with quality labels that can be computationally accessed. The evidence dynamic monitoring and updating unit runs automatically periodically. It initiates searches in external medical literature databases to query newly published literature related to consumables in the evidence-based evidence base. When new evidence text is discovered, this unit triggers the same processing flow as the data fusion module and the automated quality rating unit, namely, performing structured extraction and quantitative rating, and writing the generated new standardized evidence item into the evidence base, realizing the automatic tracking and integration of medical knowledge progress.

[0064] The evidence version management and traceability unit maintains a complete historical record of evidence. Whenever an evidence entry corresponding to a consumable is updated, whether due to the addition of new evidence or a rating adjustment, this unit does not simply overwrite the old entry. Instead, it saves the old entry as a historical version and records the time, triggering reason, and content summary of the update, forming a time-ordered evidence evolution chain for that consumable. Among these technical means, the automated quality rating unit uses machine-executable rules and quantitative scoring tables to transform subjective, qualitative expert judgments into objective, quantitative machine scores. This provides key input parameters for subsequent personalized analysis modules to perform weighted calculations or credibility screening based on evidence quality. The evidence dynamic monitoring and update and version management traceability unit ensures the timeliness and auditability of the knowledge base. Its function is to enable the system to simulate the continuous learning process of human experts and meet the compliance requirements for process traceability in medical decision-making. Therefore, the entire evidence-based management module, through the synergy of algorithmic rules and data management features, systematically solves the technical problems of inconsistent quality, lagging updates, and version chaos in the application of evidence-based medicine evidence. It constructs a high-quality, evolvable, and traceable digital evidence base, enabling subsequent personalized analysis to be built upon solid and dynamically updated evidence-based knowledge. For example, regarding "biological artificial hip joint prostheses," the system not only stores multiple research data on their five-year survival rates but also includes a quality rating for each data point. It also fully records how the evidence base updates its overall efficacy assessment and credibility after a new long-term follow-up (RWD) study is published. This provides clinicians with a clear, tiered overview of evidence when dealing with specific patients.

[0065] Furthermore, the system also includes a personalized analysis module, which is connected to the data fusion module and the evidence-based management module respectively. It is used to retrieve matching evidence from the evidence-based evidence library for the surgical scenario corresponding to the target patient's diagnosis and treatment data, combined with the patient's willingness to use consumables in the diagnosis and treatment data, and output multi-dimensional analysis results for multiple optional consumables through a pre-established patient-surgical scenario-consumable matching model.

[0066] Specifically, the personalized analysis module includes: a feature and intention processing unit, which extracts patient treatment data from the structured dataset and quantifies individual information, surgical scenario parameters, and patient consumable usage intention data from the patient treatment data into a patient-surgical scenario feature vector and a preference weight vector; a multi-dimensional prediction unit, which retrieves evidence element data matching candidate consumables from the evidence-based evidence library based on the patient-surgical scenario feature vector, and outputs analysis results including efficacy prediction, risk prediction, and cost-effectiveness prediction through a pre-established patient-surgical scenario-consumable matching model; and a score synthesis unit, which converts the analysis results of efficacy prediction, risk prediction, and cost-effectiveness prediction into scores according to standardized rules, and performs weighted fusion with the preference weight vector to generate multi-dimensional analysis results including clinical value score, economic value score, and comprehensive recommendation index.

[0067] More specifically, the personalized analysis module begins with the feature and intention processing unit. This unit extracts medical data associated with the target patient from the structured dataset provided by the data fusion module. This data already includes standardized individual information and surgical scenario parameters. The patient's individual information and surgical scenario parameters are then fused and encoded. For continuous variables such as age and body mass index, standardization is performed. For categorical variables such as gender, surgical method, and anesthesia grade, one-hot encoding or word embedding techniques are used to convert them into numerical vectors. Risk factors such as diabetes and smoking history are represented as binary or ordered numerical values. All processed features are concatenated in a preset order to form a unified feature vector. Simultaneously, this unit processes patient consumable usage intention data, which can come from structured questionnaires used in preoperative assessments. The patient's level of attention to dimensions such as cost sensitivity to surgical consumables, tolerance for potential risks, and preference for recovery speed is transformed into a preference weight vector, with the sum of the weights for each dimension being a fixed value, thereby quantifying the patient's individual value orientation.

[0068] Subsequently, the multi-dimensional prediction unit is triggered. This unit receives the patient-surgical scenario feature vector x and the patient preference weight vector ω. For each candidate consumable, the system retrieves the set of evidence entries Z associated with it from the evidence-based evidence base. m Based on clinical pathways or initial screening by physicians, several candidate consumable identifiers are identified. For each candidate consumable, the multi-dimensional prediction unit queries the evidence-based management module to retrieve all associated evidence element data and their corresponding final quality ratings. This evidence data, along with the patient-surgical scenario feature vector, is input into a pre-established patient-surgical scenario-consumable matching model.

[0069] The pre-established patient-surgical-consumable matching model contains three functionally coupled sub-models: a efficacy prediction sub-model, a risk prediction sub-model, and a cost-benefit prediction sub-model. The efficacy prediction sub-model is configured to: use the average efficacy data of the target consumable in the evidence-based database and the feature vector of the study population as a benchmark; calculate the difference between the patient-surgical-scenario feature vector and the feature vector of the study population; input the difference into a trained efficacy adjustment model to output an efficacy adjustment amount; and then combine the efficacy adjustment amount with the average efficacy data of the population to obtain the individualized efficacy prediction value for the patient. For example, the efficacy prediction sub-model uses evidence entries Z retrieved from the evidence-based database that are associated with the target consumable. m For input, each piece of evidence includes population average efficacy data. and standardized research population feature vectors For the current patient, the feature vector of the current patient-surgical scene is calculated by weighted Euclidean distance, multiplied by the feature vector of each study population. Difference between To establish a mapping model between feature variability and changes in treatment efficacy, historical patient treatment data was used to train the model by comparing the actual outcomes of surgeries using corresponding consumables with their evidence-based predictions. The training objective was to understand the influence of feature variability on the deviation of actual treatment efficacy from the group mean. The mapping model received... And output a corresponding therapeutic adjustment amount. The mathematical formula for the individualized efficacy prediction sub-model constructed in this embodiment can be expressed as:

[0070] ,

[0071] In the formula, The system predicts key efficacy indicators for the current target patient when using candidate consumable m. For the first The combined weight of each piece of evidence is determined by the first piece of evidence. The final quality rating of the evidence and the first The degree of matching between the research population and the characteristics of current patients is jointly determined by the evidence; Representing the The group average efficacy data of the evidence; Representing the The efficacy adjustment based on the evidence.

[0072] In the mathematical calculation formula of the individualized efficacy prediction sub-model constructed above, for each piece of evidence, the efficacy adjustment model is used to adjust for differences in patient and study characteristics. The average efficacy of the group was quantitatively corrected to obtain... Each piece of evidence is assigned a combined weight based on its quality and relevance to the patient. The efficacy contribution values ​​after all evidence adjustments are weighted and averaged according to their respective combination weights to obtain the final individualized efficacy prediction value. .

[0073] The efficacy prediction sub-model uses the average functional improvement data of the population corresponding to each prosthesis in the evidence-based evidence library and the feature vector of the study population as a benchmark, and combines the patient-surgical scenario feature vector difference to output individualized efficacy prediction values. The closed-loop feedback traceability module collects the patient's KSS improvement as the actual follow-up efficacy during the postoperative follow-up window, and pairs it with the efficacy prediction value to form a calibration data pair. As shown in Figure 2, the consistency verification diagram between the efficacy prediction results and the real-world follow-up efficacy in the patient-surgical scenario-consumable matching model shows that the scatter points of each case are close to the ideal consistency line, indicating that under this consumable selection scenario, the individualized efficacy prediction output by the system is consistent with the real-world follow-up efficacy, and the scatter points are close to the y=x line, thus supporting the generation of clinical value scores for candidate consumables and collaborative decision-making between doctors and patients.

[0074] Furthermore, the risk prediction sub-model is configured to: identify patient-specific risk-enhancing factors based on the patient-surgical scenario feature vector, and input the baseline risk data of the target consumables in the evidence-based database into the risk prediction model, outputting the individualized conditional probability of a preset adverse event; the cost-benefit prediction sub-model is configured to: simulate different clinical outcome paths based on the individualized efficacy prediction value and the individualized conditional probability, calculate the expected total medical cost by combining the procurement cost of the target consumables, related surgical and anesthesia costs, and adverse event handling costs, and compare the expected total medical cost with the health output gain to output the incremental cost-benefit ratio. For example, the system identifies and extracts a set of risk factors R(x) related to a specific adverse event from the patient-surgical scenario feature vector x according to predefined medical rules. These factors may include binary markers (such as the presence of diabetes), ordered classifications (such as ASA anesthesia classification), or continuous variables (such as preoperative white blood cell count). For example, for the risk of "surgical site infection," the extracted factors may include history of diabetes, BMI, surgical duration, and preoperative albumin level. For candidate consumables m and target adverse event type t (such as periprosthetic fracture), the model retrieves relevant evidence from an evidence-based database. This evidence provides the baseline risk of the consumable in a specific study population. When multiple studies exist, a fusion method similar to "quality weighting and similarity" used for efficacy prediction can be employed to calculate a comprehensive baseline risk estimate. In this embodiment, the mathematical formula for the constructed individualized efficacy prediction sub-model can be expressed as:

[0075] ,

[0076] In the formula, This represents the conditional probability that the current target patient will experience a specific type of adverse event t while using candidate consumable m; This represents a risk prediction function for adverse event type t, which can be trained using a random forest mathematical model, and whose internal parameters are learned from historical patient data. This represents the set of risk factors extracted from patient characteristics; This represents the baseline risk rate of consumable m for adverse event t.

[0077] The mathematical calculation formula of the personalized efficacy prediction sub-model constructed above is obtained by training... Based on patterns learned from a large number of historical cases, the model determines when a patient possesses... When describing the risk characteristics, the baseline risk of consumables. How should it be enlarged, reduced, or left unadjusted?

[0078] The cost-benefit prediction sub-model is configured to: simulate different clinical outcome paths based on the individualized efficacy prediction value and the individualized conditional probability; calculate the expected total medical cost by combining the target consumable procurement cost, related surgical and anesthesia costs, and adverse event handling costs; and compare the expected total medical cost with the health output gain to output the incremental cost-benefit ratio. Specifically, the cost-benefit prediction sub-model forms a data dependency relationship with the efficacy prediction sub-model and the risk prediction sub-model in the personalized analysis module, and together they constitute the economic evaluation link of the patient-surgical scenario-consumable matching model. During operation, it receives the individualized efficacy prediction value output by the efficacy prediction sub-model and the preset individualized conditional probability of adverse events output by the risk prediction sub-model, and establishes a set of clinical outcome paths corresponding to the target consumable based on these two types of outputs to achieve outcome-driven cost extrapolation. The clinical outcome pathway uses the surgical scenario as a constraint, divides the key outcomes after the application of consumables into multiple enumerable states, and establishes the transition relationship between states. The states include at least the following situations: efficacy target achieved and no adverse events occurred, efficacy target achieved but adverse events occurred, efficacy target not achieved leading to re-intervention or prolonged hospitalization, and adverse events leading to additional treatment. The probability of occurrence of each state is jointly determined by the individualized efficacy prediction value and the individualized conditional probability of adverse events, and a path probability set that can be used for expected value calculation is formed through probability normalization.

[0079] Cost elements are provided in a structured record format within the data fusion module, including the procurement cost of target consumables, surgical and anesthesia-related expenses, and project-based expenses related to the management of pre-set adverse events. Adverse event management costs are obtained through a cost dictionary or historical case cost statistics and correlated with adverse event types. The costs of re-intervention or extended hospitalization are derived by combining inpatient day costs, re-operation costs, or drug and consumable costs. Based on the correspondence between path probabilities and cost elements, the sub-model generates path costs for each clinical outcome path and aggregates them to obtain the expected total medical cost. This couples cost assessment with patient risk status, surgical scenario differences, and consumable adverse event risks, addressing the problem of disconnect between cost evaluation and clinical outcomes in existing technologies, leading to non-transferable conclusions. The data caliber maintains consistency between health output gain and efficacy prediction results. By mapping individualized efficacy prediction values ​​to the degree of efficacy improvement or quality of life improvement and combining this with the follow-up period, health output indicators are obtained. This achieves cost-effectiveness comparability on the same time scale. Based on this, the difference in expected total medical costs between the target consumables and control consumables is compared with the difference in health output gain to output the incremental cost-benefit ratio. In an example scenario, if the procurement costs of two candidate implantable consumables differ in the same orthopedic replacement surgery scenario, and one of them has a higher probability of infection in elderly patients with comorbidities, the increased conditional probability of infection in the risk prediction output will trigger an increase in the probability of infection treatment pathways and add corresponding treatment costs, thereby increasing the expected total medical costs and reflecting unfavorable changes in the incremental cost-benefit ratio. This provides an economic basis consistent with the patient's risk characteristics for collaborative decision-making between doctors and patients and improves the scientificity, interpretability, and auditability of consumable selection.

[0080] After obtaining the analysis results of efficacy prediction, risk prediction, and cost-benefit prediction, the scoring synthesis unit converts these heterogeneous predicted values ​​into comparable scores under the same dimension according to preset standardized mapping rules. For example, for efficacy prediction, the system maps it to the interval of 0 to 100 through a piecewise linear function to generate a clinical value score, where a higher score indicates better expected efficacy. For risk prediction, the system summarizes the individualized probabilities of major adverse events and converts them into a safety dimension score through an inverse mapping function, with high risk corresponding to a low score. For cost-benefit prediction, the system compares the incremental cost-benefit ratio with a preset cost-effectiveness threshold and assigns a corresponding level or value based on its interval (e.g., below the threshold, within the threshold interval, above the threshold), generating an economic value score.

[0081] After obtaining the aforementioned basic scores, this unit introduces a preference weight vector generated by the feature and intention processing unit. This vector quantifies the patient's emphasis on dimensions such as efficacy, safety, and cost. The score synthesis unit performs a weighted fusion calculation, multiplying the clinical value score, safety dimension score, and economic value score by their corresponding preference weights, and summing the products to obtain a comprehensive recommendation index. For example, if a patient is highly concerned about long-term efficacy and insensitive to cost, the weight of the clinical value score will be set higher, while the weight of the economic value score will be correspondingly reduced, making the final comprehensive recommendation index more inclined to reflect efficacy advantages. In this process, the design of the standardized mapping rule incorporates clinical value judgments and health economics consensus, and its function is to transform technical predictions into evaluation indicators with clear clinical significance. The weighted fusion algorithm mathematically integrates objective technical evaluation with subjective patient value orientation, and its function is to unify the optimal technical solution and the patient preference solution, jointly solving the technical problem of difficulty in forming a single, operable recommendation conclusion under multiple value objectives. Therefore, through standardization and weighted fusion, the system ultimately outputs a set of final analysis results for each candidate consumable, which includes clinical value score, economic value score and comprehensive recommendation index. This results are both objective and quantitative, and reflect individual value, providing a clear and comparable basis for decision-making interaction.

[0082] Furthermore, the system also includes a decision interaction module connected to the personalized analysis module. This module generates structured decision options based on the multi-dimensional analysis results and supports collaborative confirmation of the target consumable plan by doctors and patients through a visual interactive interface, recording the confirmation results. Specifically, the decision interaction module includes: a decision option generation unit, used to encapsulate each candidate consumable into a structured decision option object based on the evidence-based evidence library and the multi-dimensional analysis results. The structured decision option object includes a comprehensive recommendation index, a summary of key advantages and disadvantages, and a final quality rating of the consumable evidence; an interface rendering unit, used to render differentiated interactive interfaces according to the user's role. The physician interface displays a comparison matrix of candidate consumables and provides a detailed view linked to efficacy, risk, cost, and evidence summary. The patient interface presents the differences in a non-professional and visual manner and provides preference input controls; and a confirmation recording unit, used to record the consumable plan jointly confirmed by doctors and patients, the confirmation time, information of participating parties, and key decision considerations.

[0083] More specifically, the decision interaction module receives multi-dimensional analysis results from the personalized analysis module, which include quantitative data such as clinical value score, economic value score, and comprehensive recommendation index for each candidate consumable. Based on these results, and combined with key evidence summaries and quality ratings of the corresponding consumables extracted from the evidence-based evidence library, the decision option generation unit encapsulates each candidate consumable into a structured decision option object. This object not only integrates the aforementioned quantitative scores but also automatically extracts key advantages and disadvantages descriptions through preset summary generation rules, such as "Consumable A has a high evidence rating for long-term durability but a higher procurement cost; Consumable B is more cost-effective but has limited evidence support for heavily active patients."

[0084] The interface rendering unit dynamically renders differentiated interactive interfaces based on the user's role information on the access terminal. When the user role is a physician, the system renders a physician interface, the core of which is a comparison matrix of candidate consumables, clearly displaying the comprehensive recommendation index, key scores, and evidence quality levels of each consumable in a table or card format. When the physician clicks on any consumable or specific indicator in the matrix, the interface retrieves and displays a detailed view related to that element, such as expanding a list of all high-quality evidence supporting the consumable, or displaying the confidence intervals for efficacy and risk prediction through charts, supporting in-depth review. When the user role is a patient, the system renders a patient interface, which avoids using technical jargon and instead translates key differences into plain language and intuitive visual charts, such as using star ratings to indicate recommendation strength and using bar charts of different lengths to compare costs and expected recovery times. Simultaneously, the patient interface embeds preference input controls, allowing patients to adjust their emphasis on dimensions such as "cost," "recovery speed," and "long-term effects" in real time via a slider. These adjustments are fed back to the system in real time and trigger a reweighting and reordering of the analysis results.

[0085] The confirmation recording unit captures the final interaction state after the doctor and patient reach an agreement through the interface. This unit records the confirmed selected consumable plan, the timestamp of plan confirmation, the identities of the participating physicians and patients, and optionally, the tags of decision-making considerations that were highlighted or discussed during the interaction. In this process, the algorithm rules for generating structured decision option objects are tightly coupled with the differentiated logic of interface rendering. Its function is to transform complex backend analysis data into decision information that can be effectively understood by users with different cognitive levels on the front end. This solves the technical problems of information asymmetry and opaque decision-making basis in traditional doctor-patient communication. Simultaneously, the structured recording function of the recording unit, combined with the aforementioned data association logic, creates a traceable digital archive for each decision instance. This archive is connected to the subsequent feedback and traceability module, forming a closed-loop foundation supporting continuous improvement and auditing of decision quality.

[0086] Furthermore, the system also includes a closed-loop feedback traceability module, which is connected to the decision interaction module and the personalized analysis module. This module is used to collect real-world clinical feedback data under the application of consumables corresponding to the confirmation result, and to align the real-world clinical feedback data with the multi-dimensional analysis results output by the personalized analysis module to form calibration data. Based on the calibration data, the patient-surgical scenario-consumable matching model is iteratively optimized.

[0087] The closed-loop feedback traceability module includes: a tracking identifier generation and feedback collection unit, used to generate a globally unique decision tracking identifier for each consumable confirmation, and collect actual efficacy, actual occurrence of adverse events, and actual cost information from the hospital information system and follow-up system based on the tracking identifier; a prediction-fact alignment unit, used to align the collected actual efficacy, actual occurrence of adverse events, and actual cost information with the prediction results output by the personalized analysis module to form calibration data pairs; and a model iteration unit, used to trigger the retraining of the matching model and update the model parameters when the calibration data pairs reach a preset number or deviation threshold.

[0088] More specifically, the closed-loop feedback traceability module is activated when the decision interaction module records the consumable confirmation result. The traceability identifier generation and feedback collection unit generates a globally unique decision traceability identifier for this confirmation, which is strongly correlated with the patient, surgery, consumable, and decision time information in the confirmation record. Based on this identifier, this unit continuously collects real-world clinical feedback data related to this consumable application from the hospital information system, electronic medical records, and patient follow-up system through data interfaces. This data includes actual efficacy indicators assessed through postoperative imaging or functional scoring, the actual occurrence and type of adverse events extracted in a structured manner from medical records or obtained from the adverse event reporting system, and actual medical cost information summarized from the billing system. The collection process ensures that the feedback data accurately corresponds to the previous decision instance.

[0089] The prediction-fact alignment unit receives the aforementioned feedback data and, based on the decision tracking identifier, retrieves the multi-dimensional analysis results generated for this decision from the historical output of the personalized analysis module. Specifically, this includes the individualized efficacy prediction, risk probability prediction, and cost-benefit prediction output at that time. This unit performs data alignment, pairing the collected actual efficacy with the predicted efficacy value, pairing the actual adverse events with the predicted risk probability, and pairing the actual total cost with the predicted expected cost value, thereby forming one or more sets of "predicted value-actual value" calibration data pairs. These calibration data pairs quantify the difference between model predictions and real-world results.

[0090] The model iteration unit continuously monitors the accumulated calibration data pairs. When the number of calibration data pairs reaches a preset batch training threshold, or when the average deviation of a specific type of prediction (such as the infection risk prediction of a certain type of consumable) exceeds a preset fault tolerance threshold, the unit automatically triggers an optimization process. This process uses all accumulated calibration data pairs as training samples to retrain the corresponding sub-models (such as efficacy prediction sub-models and risk prediction sub-models) in the patient-surgical scenario-consumable matching model. During retraining, the model parameters are updated based on the objective of minimizing prediction error. After training, the new version of the model parameters is deployed for subsequent personalized analysis, thus completing a complete decision-feedback-optimization closed loop. In this technology chain, the generation and association mechanism of decision tracking identifiers is a core technical feature. Its function is to establish an immutable and traceable causal relationship chain for each decision and its subsequent results in a complex medical data environment. This solves the technical problem that real-world data is difficult to accurately backtrack and align with specific decisions. The algorithm logic of prediction-fact alignment relies on the standardized data structure output by the preceding module. Its function is to transform unstructured clinical outcomes into standardized supervision signals that can be used for model training. Therefore, the closed-loop feedback traceability module, through a combination of techniques including identifier association, data alignment, and triggered retraining, enables the system to continuously optimize its core prediction model using the application feedback it generates. This achieves the evolution from a static auxiliary system to an intelligent system with self-learning and continuous improvement capabilities, effectively solving the problem of predictive performance degradation of machine learning models in the medical field due to changes in clinical practice and data drift. For example, when the system finds that its prediction of the efficacy of a novel biomaterial in elderly patients remains optimistic but the actual follow-up results are weak, the calibration data accumulated through closed-loop feedback can drive the efficacy adjustment model to correct the mapping relationship between the characteristic variability of this population and the efficacy adjustment amount, making subsequent predictions more accurate.

[0091] In addition, the closed-loop feedback traceability module also includes a report solidification unit, which is used to solidify and generate cryptographic hash values ​​based on the multi-dimensional analysis results, the patient-surgical scenario-consumable matching model version, the content recorded by the confirmation record unit, and the content generated and collected by the tracking identifier generation and feedback collection unit, based on the actual consumables selected by the specific patient, and encapsulate and output them in a preset audit format to support a full-link traceability evaluation report for third-party integrity verification.

[0092] Specifically, the report solidification unit is activated after the consumables plan is confirmed and a preset follow-up period has elapsed, aiming to generate an immutable end-to-end decision audit report. This unit aggregates snapshots of all key data related to the decision from various modules of the system, based on decision tracking identifiers. This data includes: multi-dimensional analysis results generated by the personalized analysis module for candidate consumables at the decision-making moment, along with the patient feature vectors upon which they are based; confirmation records recorded by the decision interaction module, including the finally selected consumables, participant information, and decision time; the specific version identifier used by the patient-surgical scenario-consumable matching model at the time of the decision; and real-world clinical feedback data collected and aligned by the closed-loop feedback traceability module, namely actual efficacy, adverse events, and cost information. The report solidification unit structures and organizes the above heterogeneous data according to chronological order and logical relationships, encoding it into a preset audit data format.

[0093] Subsequently, the unit performs a cryptographic hash operation on the complete structured data block, such as using the SHA-256 algorithm, to generate a unique hash value. This hash value, as a digital fingerprint of data integrity, is embedded in the final audit report. The report itself is packaged in a combination of machine-readable and human-readable formats, such as using JSON-LD format to preserve semantics, supplemented by a PDF summary view. The report explicitly includes a tracking identifier for the decision, references or summaries of all data snapshots, and key timestamps and version numbers. Any subsequent tampering with the report content, even a change of a single character, will result in a mismatch between the recalculated hash value and the original embedded value, making it easily detectable by the verification process.

[0094] In this process, the cryptographic hashing algorithm executed by the report solidification unit and the aggregation logic of the data snapshot function support and interact with each other. The data aggregation logic ensures the complete capture of information from all key stages throughout the decision-making lifecycle, and its function is to construct an auditable and coherent chain of facts. The cryptographic hashing algorithm, on the other hand, applies an unforgeable integrity seal to this chain of facts, and its function is to logically bind physically dispersed data into a cryptographically protected whole of evidence. This combination of technologies jointly solves the technical problems of insufficient transparency, auditability, and evidence credibility in the decision-making process of medical AI decision support systems when facing regulatory reviews, quality assessments, or dispute resolution. The generated report enables third parties (such as hospital quality control departments, medical insurance auditors, or medical device regulatory agencies) to independently verify, without relying on the system backend, whether a consumable recommendation decision was objectively generated based on the effective model and data at the time, and whether the subsequent real-world results were truthfully recorded and fed back. This transforms the "black box" decision-making process of the intelligent system into a verifiable and traceable "white box" chain of evidence, enhancing the compliance, credibility, and accountability of the entire system. For example, the report can be used to demonstrate that the selection of a high-priced consumable was based on the latest high-quality evidence, personalized analysis of patient characteristics, and a clear joint decision-making process between doctors and patients, rather than other non-clinical factors, providing reliable technical evidence for the practice of value-based healthcare.

[0095] It should be further explained that the data fusion module, evidence-based management module, personalized analysis module, and decision interaction module provided in this embodiment are each specifically deployed in a hardware environment consisting of servers, storage devices, and network devices, and their specific functions are implemented through the execution of software instructions. Specifically, the data fusion module is deployed in the hospital's data integration server or medical information platform server. It communicates with the databases of the hospital's information system and supply chain management system through the standard data interface configured on the server and performs network data collection tasks. Its data standardization and association construction functions are implemented by the corresponding data processing program running on the server, and the generated structured dataset is stored in a centralized storage device or distributed database cluster connected to it. The evidence-based management module and personalized analysis module, as core computing units, are deployed on the hospital's application server or a virtual computing instance of the cloud computing platform. This server or instance carries the evidence quality rating algorithm, the patient-surgical scenario feature vector calculation program, and the patient-surgical scenario-consumable matching model. The training and iterative optimization process of the matching model itself is usually completed on a training server or cloud computing GPU instance with stronger computing power. The trained model parameters are then deployed to the aforementioned application server for real-time prediction. The evidence-based database, as a dynamically updated knowledge base, is stored in a relational or non-relational database that is directly associated with the aforementioned application server.

[0096] The decision-making interaction module exists as a front-end application. Its interface rendering unit runs on web browsers or dedicated client applications on physician workstations, mobile terminals, or ward interactive whiteboards, exchanging data with the back-end application server via the hospital intranet or a secure network channel. The logic for generating and confirming decision options is supported by service programs on the back-end application server. The closed-loop feedback traceability module is distributed across multiple parts of the system. Its tracking identifier generation, feedback collection, and prediction-fact alignment functions are implemented by a background service program deployed on the application server, responsible for scheduling data collection and processing tasks. The model iteration unit involves retraining, and its triggering and execution environment is the same as the model training environment, located on the training server. The report solidification unit can be deployed on the application server or a dedicated report generation server, responsible for aggregating data, performing hash calculations, and formatting report output. All modules communicate and exchange data securely through the hospital's intranet or virtual private cloud, collectively forming a complete physical system deployed on specific hardware devices and operating collaboratively through software.

[0097] Through the system architecture and data processing flow described in the above embodiments, the intelligent evaluation of medical consumables forms a workable closed loop in the surgical scenario corresponding to the target patient: The data fusion module standardizes and encodes patient diagnosis and treatment data, surgical scenario parameters, and consumable usage and cost data from the hospital information system, supply chain management system, and follow-up system, and associates them with unique consumable identifiers. Simultaneously, it structures external medical literature evidence into evidence elements and writes them into the evidence-based evidence library along with the evidence quality rating results, thereby providing consistent data standards and traceable data sources for subsequent calculations; the personalized analysis module is based on patient-surgical scenario feature vectors, preference weight vectors, and evidence-based evidence. The evidence element data corresponding to candidate consumables in the database are used by a matching model to output efficacy prediction, risk prediction, and cost-effectiveness prediction, which are then synthesized into clinical value scores, economic value scores, and a comprehensive recommendation index. This transforms consumable evaluation from a group average conclusion into an individualized result that matches the patient's risk status, surgical complexity, and preferences. The decision interaction module encapsulates the analysis results into structured decision options and supports collaborative confirmation between doctors and patients, enabling the evaluation basis to be presented and recorded in an understandable form. The closed-loop feedback traceability module aligns actual clinical outcomes with predicted results to form calibration data and uses it for model iteration, ensuring the system maintains stability and applicability in continuous application. Thus, this embodiment achieves unified quantification and scenario-based application of evidence-based evidence and real-world data at the scientific level; at the interpretability level, it enables the reproducibility of the link from input features and evidence elements to score output; and at the auditability level, it enables full traceability from data source, evidence quality, model version to decision record. This solves the technical problems of strong subjectivity, difficulty in individualization, low evidence utilization efficiency, and lack of closed-loop optimization mechanisms in consumable evaluation, and provides reusable technical support in clinical consumable selection, doctor-patient communication, and management decision-making scenarios.

[0098] The above description is merely a specific embodiment of this specification. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the scope of protection of this specification is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this specification, and these modifications or substitutions should all be covered within the scope of protection of this specification.

Claims

1. A medical consumables intelligent evaluation system integrating evidence-based medicine and personalized analysis, characterized in that, include: The data fusion module is used to obtain patient diagnosis and treatment data, consumable information and corresponding evidence-based medicine evidence from internal and external data sources of the hospital, and generate structured datasets; The evidence-based evidence management module is used to quantify and rate the evidence-based medical evidence in the structured dataset according to preset evidence quality assessment rules, and to build an evidence-based evidence library associated with consumable identification. The personalized analysis module is used to retrieve matching evidence from the evidence-based database for surgical scenarios corresponding to the target patient's medical data, combined with the patient's willingness to use consumables in the medical data, and output multi-dimensional analysis results for multiple optional consumables through a pre-established patient-surgical scenario-consumable matching model. The decision interaction module is used to generate structured decision options based on the multi-dimensional analysis results, and supports doctor-patient collaborative confirmation of the target consumable plan and records the confirmation results through a visual interactive interface. The closed-loop feedback traceability module is used to collect real-world clinical feedback data under the application of consumables corresponding to the confirmation results, align the real-world clinical feedback data with the multi-dimensional analysis results to form calibration data, and iteratively optimize the patient-surgical scenario-consumable matching model based on the calibration data. The patient-surgical scenario-consumable matching model includes an efficacy prediction sub-model, a risk prediction sub-model, and a cost-benefit prediction sub-model; wherein, the efficacy prediction sub-model is configured to: use the evidence-based database... Using the average efficacy data of the target consumable and the feature vector of the study population as a benchmark, the difference between the patient-surgical scenario feature vector and the feature vector of the study population is calculated. The difference is then input into a trained efficacy adjustment model to output an efficacy adjustment amount. The efficacy adjustment amount is then combined with the average efficacy data of the group to obtain the individualized efficacy prediction value for the patient. A risk prediction sub-model is configured to: identify patient-specific risk enhancement factors based on the patient-surgical scenario feature vector, and input the baseline risk data of the target consumable in the evidence-based library into the risk prediction model to output the individualized conditional probability of a preset adverse event. A cost-benefit prediction sub-model is configured to: simulate different clinical outcome paths based on the individualized efficacy prediction value and the individualized conditional probability, calculate the expected total medical cost by combining the procurement cost of the target consumable, related surgical and anesthesia costs, and adverse event handling costs, and compare the expected total medical cost with the health output gain to output the incremental cost-benefit ratio.

2. The intelligent evaluation system for medical consumables integrating evidence-based medicine and personalized analysis according to claim 1, characterized in that, The data fusion module includes: a multi-source data acquisition unit, used to acquire patient diagnosis and treatment data and consumable circulation data from hospital information systems and supply chain management systems through standard data interfaces, and to acquire evidence-based medical evidence text related to the target consumable from external medical literature databases through a network acquisition mechanism; a data standardization unit, used to map diagnosis, surgery, and consumable names to standard codes based on a medical terminology mapping table, and to perform named entity recognition and relation extraction on unstructured text to generate structured records; an association construction unit, used to associate the same consumable's unique identifier with the user's, surgical scenario, cost data, and clinical feedback data to construct a patient-surgical scenario-consumable association data chain; and a fusion storage unit, used to store the structured records and the association data chain.

3. The intelligent evaluation system for medical consumables integrating evidence-based medicine and personalized analysis according to claim 2, characterized in that, The structured records generated by the data standardization unit include in-hospital patient diagnosis and treatment data and evidence element data. The patient diagnosis and treatment data includes individual patient information, preoperative assessment data, surgical scenario parameters, perioperative diagnosis and treatment data, postoperative follow-up data, and cost and resource usage data. Individual information data includes age, gender, body mass index, and underlying diseases. Preoperative assessment data includes the patient's imaging and physiological examination results, preoperative risk assessment conclusions, and willingness to use consumables. Surgical scenario parameters include the type, name, method, grade, and duration of the surgery. The evidence element data is automatically extracted from evidence-based medicine text and includes study design type, study subject characteristics, intervention measures, control measures, key outcome indicators, and corresponding statistical data. Based on this, a standardized set of evidence elements containing randomized controlled trial data and real-world data is formed.

4. The intelligent evaluation system for medical consumables integrating evidence-based medicine and personalized analysis according to claim 3, characterized in that, The evidence-based management module includes: an automated quality rating unit, used to perform quantitative quality assessments on randomized controlled trial data and real-world data respectively, wherein: for randomized controlled trial data, key outcome indicators are assessed based on machine-executable evidence rating rules in the dimensions of risk of bias, inconsistency, indirectness, imprecision, and publication bias to generate a preliminary quality rating; for real-world data, a quantitative scoring table is used to score data from the dimensions of data source, study design, bias control, and consistency, and the scores are normalized to obtain a credibility weight; and a final quality rating for consumable evidence is generated based on the preliminary quality rating and credibility weight; and an evidence library construction unit, used to encapsulate the final quality rating and corresponding evidence element data into standardized evidence items and store them in the evidence-based evidence library.

5. The intelligent evaluation system for medical consumables integrating evidence-based medicine and personalized analysis according to claim 4, characterized in that, The evidence-based management module also includes: an evidence dynamic monitoring and update unit, which is used to periodically search external medical literature databases and obtain new evidence texts related to consumables in the evidence-based evidence library, trigger the structured processing and final quality rating of the new evidence, and write it into the evidence-based evidence library; and an evidence version management and traceability unit, which is used to manage the version of evidence entries corresponding to the same consumable, retain historical versions and record evidence update logs to form a traceable evidence evolution chain.

6. The intelligent evaluation system for medical consumables integrating evidence-based medicine and personalized analysis according to claim 1, characterized in that, The personalized analysis module includes: a feature and intention processing unit, which extracts patient diagnosis and treatment data from the structured dataset, and quantifies individual information, surgical scenario parameters, and patient consumable usage intention data from the patient diagnosis and treatment data into a patient-surgical scenario feature vector and a preference weight vector; a multi-dimensional prediction unit, which retrieves evidence element data matching candidate consumables from the evidence-based evidence library based on the patient-surgical scenario feature vector, and outputs analysis results including efficacy prediction, risk prediction, and cost-effectiveness prediction through a pre-established patient-surgical scenario-consumable matching model; and a score synthesis unit, which converts the analysis results of efficacy prediction, risk prediction, and cost-effectiveness prediction into scores according to standardized rules, and performs weighted fusion with the preference weight vector to generate multi-dimensional analysis results including clinical value score, economic value score, and comprehensive recommendation index.

7. The intelligent evaluation system for medical consumables integrating evidence-based medicine and personalized analysis according to claim 1, characterized in that, The decision-making interaction module includes: a decision option generation unit, used to encapsulate each candidate consumable into a structured decision option object based on the evidence-based evidence library and the multi-dimensional analysis results. The structured decision option object includes a comprehensive recommendation index, a summary of key advantages and disadvantages, and a final quality rating of the consumable evidence; an interface rendering unit, used to render differentiated interactive interfaces according to the user's role. The physician interface displays a comparison matrix of candidate consumables and provides a detailed view linked to efficacy, risk, cost, and evidence summary. The patient interface presents the differences in a non-professional and visual manner and provides preference input controls; and a confirmation recording unit, used to record the consumable plan, confirmation time, information of participating parties, and key decision-making considerations jointly confirmed by the physician and patient.

8. The intelligent evaluation system for medical consumables integrating evidence-based medicine and personalized analysis according to claim 7, characterized in that, The closed-loop feedback traceability module includes: a tracking identifier generation and feedback collection unit, used to generate a globally unique decision tracking identifier for each consumable confirmation, and collect actual efficacy, actual occurrence of adverse events, and actual cost information from the hospital information system and follow-up system based on the tracking identifier; a prediction-fact alignment unit, used to align the collected actual efficacy, actual occurrence of adverse events, and actual cost information with the prediction results output by the personalized analysis module to form calibration data pairs; and a model iteration unit, used to trigger the retraining of the matching model and update the model parameters when the calibration data pairs reach a preset number or deviation threshold.

9. The intelligent evaluation system for medical consumables integrating evidence-based medicine and personalized analysis according to claim 8, characterized in that, The closed-loop feedback traceability module further includes a report solidification unit, which is used to solidify and generate cryptographic hash values ​​based on the multi-dimensional analysis results, the patient-surgical scenario-consumable matching model version, the content recorded by the confirmation record unit, and the content generated and collected by the tracking identifier generation and feedback collection unit, based on the actual consumables selected by the specific patient, and encapsulate and output the report in a preset audit format to support a full-link traceability evaluation report for third-party integrity verification.

Images