Total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology

Abstract

The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology includes intelligent TPN formula module, pharmaceutical intelligent review module, intelligent preparation and quality control module, intelligent infusion management module, real-time dynamic monitoring module and full-process closed-loop management module; from prescription optimization to pharmaceutical review, from preparation quality control to infusion management, from dynamic monitoring to closed-loop management, each link is seamlessly connected and linked together, and the data flow is transferred between modules, driving the full-process intelligent closed-loop control of TPN treatment. The intelligent formula is continuously iterated and optimized based on the results of pharmaceutical review, and pharmaceutical review continuously improves the rule base based on this; dynamic monitoring triggers real-time adjustment of infusion control, and the massive data of the infusion process feeds back to the continuous learning of the monitoring model; closed-loop management promotes continuous improvement of each link through full-process data analysis, and the improvement results of each link are reflected in the data, forming a virtuous circle.

Description

Technical Field

[0001] The present invention relates to the technical field of total parenteral nutrition management and control systems, and more specifically, to a total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technologies. Background Art

[0002] Total parenteral nutrition (TPN) is an important means of providing nutritional support for patients who are unable to eat orally or have intestinal insufficiency. However, due to the complex formula of TPN, a wide variety of nutrients, and a high risk of metabolic complications, there are still many challenges in the clinical implementation process.

[0003] Traditional TPN treatment mainly relies on the doctor's experience to design and adjust prescriptions, lacks individual considerations, and is difficult to accurately assess and meet the nutritional needs of patients, resulting in insufficient or excessive nutritional support, which affects the efficacy. Domestic and foreign studies have shown that the incidence of malnutrition in patients with empirical TPN regimens is as high as 30%-50%. At the same time, intravenous hypernutrition and electrolyte disorders can also cause metabolic complications such as hyperglycemia and hyperkalemia, with an incidence of 20%-30%.

[0004] In recent years, the application of tools and consensus such as the National Nutrition Risk Screening System (NRS2002) and the Society for Parenteral and Enteral Nutrition (ASPEN) guidelines have guided the standardized implementation of TPN to a certain extent, but their ability to dynamically manage patients' nutritional status and metabolic risks is limited. Studies have shown that even under the guidance of the NRS2002 and ASPEN guidelines, the incidence of TPN-related metabolic complications is still 10%-20%.

[0005] On the other hand, due to the complex preparation and infusion process of TPN solution, which involves many links and personnel, there is a high risk of drug mismatch, improper infusion speed, and line infection. There are about 250,000 cases of nosocomial catheter-related bloodstream infection (CRBSI) in the United States each year, causing great harm to patients. However, there is still a lack of effective control measures for the entire process of TPN implementation at home and abroad.

[0006] In addition, clinical TPN-related data are numerous, covering prescriptions, dispensing, infusion and other links, but the data are scattered and heterogeneous, lacking systematic integration and analysis, making it difficult to form a complete data chain, and unable to quantitatively evaluate and continuously optimize the treatment process. An innovative intelligent management and control system is urgently needed to collect TPN full-process data, intelligently assist decision-making during the treatment process, accurately and dynamically regulate and provide risk warnings, so as to improve the accuracy and safety of clinical nutrition therapy.

[0007] In summary, the traditional TPN treatment model still has many deficiencies in prescription formulation, treatment management, and risk control, resulting in poor patient benefits and high risk of complications. An innovative, intelligent, precise, and fully controllable TPN management model is urgently needed to accurately assess patient nutritional needs, dynamically optimize treatment strategies, standardize implementation processes, and control complication risks, thereby comprehensively improving the clinical efficacy and safety of TPN and benefiting patients. This is exactly the research and development background and urgent need of this "intelligent TPN full-process control system". Summary of the invention

[0008] 1. Technical problems to be solved

[0009] In view of the problems existing in the prior art, the purpose of the present invention is to provide a total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technologies.

[0010] 2. Technical solution

[0011] To solve the above problems, the present invention adopts the following technical solutions:

[0012] The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology includes intelligent TPN formula module, pharmaceutical intelligent review module, intelligent preparation and quality control module, intelligent infusion management module, real-time dynamic monitoring module and full-process closed-loop management module;

[0013] The intelligent TPN prescription module is used to intelligently generate personalized TPN prescriptions based on the patient's clinical data;

[0014] The pharmaceutical intelligent review module is used to review the rationality of TPN prescriptions, identify inappropriate prescriptions, and provide revision suggestions;

[0015] The intelligent preparation and quality control module is used to automatically generate preparation instructions based on the reviewed TPN prescription and perform real-time quality monitoring of the preparation process;

[0016] The intelligent infusion management module performs precise control and real-time safety monitoring of the TPN infusion process to ensure the accuracy and safety of the infusion process;

[0017] The real-time dynamic monitoring module continuously monitors the patient's vital signs, biochemical indicators, and metabolic status in real time, and dynamically adjusts the TPN treatment plan accordingly;

[0018] The full-process closed-loop management module is used for data collection, storage, sharing and tracing of the entire TPN treatment process, forming a data closed loop and promoting continuous optimization.

[0019] This intelligent TPN full-process control system fully applies artificial intelligence technology, with "closed-loop quality control" and "continuous optimization" as the core concepts, seamlessly connecting key links such as clinical decision-making, prescription review, drug preparation, infusion configuration, infusion management, and nutritional monitoring to form a highly integrated, intelligent and collaborative integrated solution. The modules in the system coordinate with each other, and strict logical relationships are set in key links to ensure the continuity of data flow and the consistency of prescription information. At the same time, the system makes full use of big data analysis and machine learning technology to continuously learn and optimize in medical data, and continuously improve the accuracy and safety of TPN treatment. We believe that the establishment and application of this system will significantly improve the hospital's TPN diagnosis and treatment level, reduce medical errors, improve patient prognosis, and be at the forefront of smart medical care. This will surely create a new model of clinical nutritional treatment and provide valuable experience for the research and development of intelligent auxiliary medical systems.

[0020] The intelligent TPN formula module first automatically extracts the patient's demographic characteristics, body composition, biochemical indicators, and diagnostic data through the electronic medical record system, and standardizes them into a format recognizable by the system; then, a pre-constructed nutritional risk assessment model is used to perform a nutritional risk score on the patient, referred to as NRS, and the NRS is used as an important reference for subsequent formula optimization; then, the standardized patient data is input into the formula optimization model, and the pharmaceutical properties of the TPN components, such as calories, protein, and electrolyte content parameters, are retrieved from the drug information database, and the sequential quadratic programming algorithm is used to solve the problem, and an optimal TPN formula prescription that takes into account both patient needs and pharmaceutical requirements is obtained. Finally, the generated TPN prescription is sent to the pharmaceutical intelligent review module for rationality review.

[0021] The pharmaceutical intelligent review module receives the TPN prescription from the intelligent TPN formula module, converts the prescription into a structured form using natural language processing technology, then retrieves the drug interaction knowledge base and the incompatibility library, and uses a rule-based reasoning algorithm to review the prescription, focusing on the following issues: whether the prescription components are consistent with the patient's condition, whether the drug dosage is appropriate, drug interactions, and incompatibility. For prescriptions that pass the review, they are directly sent to the intelligent dispensing and quality control module; for prescriptions that fail the review, the system will mark the specific problems and give revision suggestions, which will be fed back to the upstream intelligent TPN formula module for optimization and iteration until a qualified TPN prescription is generated.

[0022] The intelligent preparation and quality control module receives the qualified TPN prescription from the pharmaceutical intelligent review module, automatically generates a standardized preparation sheet, and transmits it to the automated preparation equipment; at the same time, according to the prescription requirements, it sends a drug collection instruction to the drug warehouse management system, and the drug warehouse management system feeds back the collected drug information to the intelligent preparation submodule for distribution and preparation; when the automated preparation equipment performs TPN configuration, various sensors are used to collect configuration process parameters in real time, and compare them with the preset quality standard thresholds, and the solution ratio, particle number, and stability key quality attributes are monitored in real time. Once an abnormal deviation is found, the system immediately issues an early warning and automatically Isolate problematic preparations and feed back abnormal information to the pharmaceutical intelligent audit module, triggering a re-examination and optimization of TPN prescriptions; automatically print labels for qualified TPN preparations, which contain patient information, drug information, preparation time, expiration date, storage conditions, preparer, and reviewer information, and are presented in the form of barcodes; load qualified finished drug solutions into smart infusion carts, write the corresponding relationship information between drug solutions and patients and infusion pumps into the control panel of the infusion carts, transport the infusion carts to the wards, and transmit a bagging completion signal to the intelligent infusion management module; upload the preparation process data to the full-process closed-loop management module for storage and analysis.

[0023] The intelligent infusion management module receives the prepared TPN solution from the intelligent preparation and quality control module. When the nurse inserts the infusion bag containing the patient's TPN solution into the infusion pump in the ward, the infusion pump automatically reads the solution information and matches it with the doctor's order by scanning the barcode on the infusion bag. If the match is successful, the infusion pump automatically extracts the infusion parameters, such as infusion speed, infusion time, and infusion volume, from the doctor's order after intelligent review, and automatically sets them. The nurse starts the infusion after checking again. During the infusion process, medical staff monitor the various sensors connected to the patient's venous catheter, such as pressure sensors, bubble sensors, and blood oxygen saturation sensors, and the relevant monitoring data is automatically uploaded to the intelligent infusion management module. If the infusion line is detected to be detached, blocked, or leaking, an alarm is issued, and corresponding measures are taken intelligently according to the type of abnormality, such as automatically pausing the infusion, adjusting the infusion speed, and notifying the nurse on duty to go and check. When the nurse is patrolling, the barcode of the infusion bag is scanned by a handheld terminal to obtain the current infusion progress and remaining time information. The results of each inspection are automatically recorded in the intelligent infusion management module; after the infusion is completed, the infusion pump is automatically stopped and the infusion record is saved, including the drug name, infusion time, infusion volume, average infusion speed, and infusion process curve for clinical query and medical accident tracing. All infusion process data will be uploaded to the full-process closed-loop management module in real time for audit tracing.

[0024] The real-time dynamic monitoring module continuously collects the patient's heart rate, blood pressure vital signs data, as well as blood sugar, electrolytes, and acid-base balance biochemical indicators through the interface with the ward monitoring equipment and the inspection system; these data will be transmitted to the intelligent infusion management module in real time to evaluate the safety of the infusion process; the real-time dynamic monitoring module presents the patient's vital signs trend data and laboratory test results in a rasterized manner, so that medical staff can grasp the patient's condition at any time; when there is an abnormal fluctuation in vital signs, or the laboratory test results exceed the preset threshold, the system pushes an alarm message to the doctor on duty and gives a reference diagnosis and treatment suggestions; at the same time, the module uses a machine learning algorithm to continuously calculate the patient's comprehensive nutritional score nPNI in the background to evaluate the patient's overall nutritional status; when nPNI is lower than the normal value, the system infers that the patient may have hypoproteinemia, anemia, inflammatory response nutrition-related problems, and adjusts the TPN prescription accordingly; the adjusted TPN prescription re-enters the review process of the intelligent TPN formula module until an optimized new prescription is generated, and the nutritional status assessment uses the following formula:

[0025] nPNI(t)=f(HGB(t), PALB(t), PCR(t),...)

[0026] nPNI(t) represents the nutritional status assessment at time t, taking into account multiple nutrition-related parameters such as hemoglobin HGB(t), serum albumin PALB(t), and protein degradation rate PCR(t);

[0027] The prescription adjustment adopts the following formula: If nPNI(t)<nPNIl,

[0028] Then TPN(t+1)=a(t)·Pr+b(t)·Ft+c(t)·Cb

[0029] When nPNI(t) is lower than the threshold nPNIl, the nutritional prescription TPN(t+1) for the next day (t+1) needs to be adjusted according to the current nPNI and TPN(t) of the previous day; here Pr, Ft, Cb represent the amount of protein, fat and carbohydrates respectively; coefficients a(t), b(t), c(t) are weight coefficients dynamically fitted according to nPNI(t) and TPN(t), which are used to adjust the proportion of various nutrients.

[0030] The full-process closed-loop management module collects all data generated in each link of TPN treatment through the hospital-wide information integration platform, including prescription data, dispensing data, infusion data, and monitoring data, and uses blockchain technology to establish a TPN full-process data chain to ensure the integrity, accuracy, immutability and traceability of the data; the full-process closed-loop management module is based on the five basic principles of 5R, namely Right patient, Right drug, Right dose, Right route, and Right time, to implement TPN full-process closed-loop management; the full-process closed-loop management module intuitively displays the patient's complete TPN treatment history in the form of a timeline, spanning multiple links such as outpatient, hospitalization, and discharge; each node of the timeline can be expanded to present key information, such as the time of doctor's prescription, prescription composition, configuration and adjustment records, drug traceability codes, infusion curves, and nutritional monitoring reports, for easy review of the entire process; each link strictly follows the "SCANNING IN, SCANNING "OUT", referred to as SISO specification, that is: scan the patient's wristband at the beginning of each key link, confirm the patient's identity before executing the doctor's order; at the end of each key link, scan the patient's wristband and the doctor's order again, and submit the execution result to form a closed loop; require two people to sign the key links of doctor's order, drug preparation, infusion configuration, and infusion pipeline verification, and the second signer must be authenticated by fingerprint recognition or iris recognition to eliminate the phenomenon of "signing on behalf of others"; each step of the operation forms a traceable electronic signature; all original documents, such as doctor's order, preparation, and infusion orders, are printed on special anti-tampering paper and affixed with a "one-time seal" with a unique traceability code printed on the seal; each document must be completely archived after execution for later reference; the embedded TPN quality continuous improvement submodule can implement T2-1 statistical process control, i.e. SPC, for the key indicators of doctor's order standardization, preparation accuracy, and infusion error rate. Based on the intelligent early warning of SPC control charts and continuous optimization, all data are stored in the central database in a unified format and standard, and shared according to the predetermined data access control strategy for use by relevant departments and management departments. The pharmacy department can access prescription and dispensing data to evaluate the rationality of prescriptions and the quality of dispensing; the nursing department can access infusion and monitoring data to evaluate the standardization and safety of infusion management; the medical management department can access the full process data to evaluate the overall diagnosis and treatment quality and efficiency. The full process closed-loop management module is also embedded with data analysis and mining tools to regularly analyze the full process data, generate quality reports, identify potential problems, and make continuous improvement suggestions, such as analyzing the doctor's prescription medication pattern, discovering unreasonable prescription patterns and standardizing them; analyzing the clinical efficacy of different prescriptions and optimizing the TPN prescription library; analyzing the relationship between different infusion strategies and the incidence of adverse events, and formulating best infusion practice specifications.

[0031] Input age A, gender G, height H, weight W; body composition: muscle mass M, fat mass F, water content H2O; biochemical indicators: albumin ALB, lymphocyte count LY, blood sugar GLU, renal function eGFR, liver function ALT AST; clinical diagnosis: primary disease D1, complications D2, surgical history SH, infection status INF;

[0032] The nutritional risk assessment model uses principal component analysis and logistic regression methods:

[0033] NRS=f(A,G,H,W,M,F,H2O,ALB,LY)=β 0 +β 1 PC 1 +β 2 PC 2 +...+β n PC r

[0034] Among them, PC i is the i-th principal component, β i is the corresponding regression coefficient; the objective function is to minimize NRS while meeting nutrient requirements:

[0035] Min INRS

[0036] st

[0037] Calorie requirement: HC (kcal / d) = BEE × AF × IF × TF

[0038] Protein requirement: PR (g / kg / d) = IBW × (1.2-2.0)

[0039] Fat energy supply ratio: 20%~30%

[0040] Carbohydrate energy supply ratio: 50%~70%

[0041] Electrolyte concentration limit: Na + (135-145mmol / L); K + (3.5-5.5mmol / L)

[0042] Maximum infusion volume: 40 mL / kg / day

[0043] Osmotic pressure limit: 900mOsm / L

[0044] Drug incompatibility: Incompatible drug pairs {(D1, D2), (D3, D4), ...}

[0045] The sequential quadratic programming algorithm is used to solve the optimal TPN formula:

[0046] TP={V,AA,LIP,GLU,Na + , K + , ...}

[0047] Among them, V is the TPN infusion volume, AA is the amino acid amount, LIP is the fat emulsion amount, and GLU is the glucose amount.

[0048] The medical order structuring adopts a natural language processing algorithm, and the free text of the TPN medical order is input. The specific steps are as follows:

[0049] (1) Word segmentation and part-of-speech tagging: Segment the medical order text into words and mark the part of speech of each word;

[0050] (2) Named entity recognition: identifying key entities such as drug name, dosage number, unit, and route of administration;

[0051] (3) Relationship extraction: extracting key attribute pairs of drug-dose and drug-administration route;

[0052] (4) Attribute normalization: Map synonyms, abbreviations, and typos to a standard vocabulary;

[0053] (5) Five-tuple filling: fill in the standard template of drug, dosage, route of administration, frequency, and course of treatment according to the extracted attribute pairs, and mark the attributes that cannot be filled as null values;

[0054] Output a structured quintuple consisting of drug, dose, route of administration, frequency, and duration of treatment.

[0055] The rule-based reasoning includes the following steps:

[0056] Knowledge base: (1) Drug interaction knowledge base: define groups of if-then rules, such as: If {diazepam, midazolam}∈prescription Then alarm "diazepam and midazolam should not be used at the same time";

[0057] (2) Incompatibility database: defined as if-then rules; for example, If {amino acid, fat emulsion}∈prescription & concentration (amino acid)>6% Then alarm "It is not suitable to use fat emulsion when the amino acid concentration is too high";

[0058] (3) Dose range library: defined as a value interval, for example, the glucose concentration for children should be between 5% and 12%;

[0059] Reasoning process: (1) Matching mode: match the structured medical instructions with the rule conditions of the knowledge base; (2) Conflict detection: check whether there is a logical conflict in the conclusion part of the rule; (3) Problem output: output the problem description and revision suggestions of the hit rule.

[0060] 3. Beneficial effects

[0061] Compared with the prior art, the advantages of the present invention are:

[0062] 1. Intelligent decision-making and precision treatment based on big data: Traditional TPN treatment plan formulation mainly relies on the experience and subjective judgment of doctors. This system uses machine learning algorithms to automatically learn and extract hidden association patterns from massive clinical data, and establish a mathematical mapping model between patient characteristics and TPN treatment effects, thereby assisting doctors in making more intelligent and precise treatment decisions. The system first applies NLP technology to semantically understand and extract elements from unstructured medical record data, and then maps it into a structured vector representation through feature engineering, and selects feature subsets that have a significant impact on nutritional risk assessment and TPN formula optimization. As a result, the system constructs an end-to-end learning framework from multidimensional patient information to nutritional risk probability and TPN formula. When a new patient receives TPN treatment, his or her various examination data are automatically entered into the framework, and the system can intelligently predict his or her nutritional risk and give targeted TPN prescription composition and dosage recommendations, greatly improving the accuracy of "differential treatment". More importantly, based on reinforcement learning technology, the system can collect changes in patients' biochemical indicators in real time during treatment, and dynamically adjust the TPN formula through objective functions (such as maximizing the patient's prognosis score). In this way, the best balance between standardization and personalization is achieved, and "treatment is based on the time" and "treatment is based on the person". With the continuous accumulation of case data, the system's prediction model is also constantly self-improving and iteratively updating, making the TPN treatment plan more and more intelligent and precise.

[0063] 2. Intelligent warning based on rule engine and knowledge graph: At present, various medical errors and adverse events in the clinical diagnosis and treatment process are emerging in an endless stream, including routine problems such as unreasonable prescriptions and contradictory medical instructions, as well as hidden risks such as drug compatibility and improper infusion speed. For doctors, it is not easy to accurately identify these risks in a large amount of drug instructions and literature. This system cleverly introduces rule-based reasoning technology, formalizes various medical specifications, drug instructions, expert consensus, etc. into a structured rule set in advance, and organizes them into a hierarchical and semantic medical knowledge graph using ontology construction technology. When a new TPN prescription is submitted, the system automatically matches it with the rules in the knowledge graph, and the reasoning engine strictly enforces the rules of drug usage, dosage, compatibility, etc., quickly judges the rationality and safety of the prescription, and gives revision suggestions for specific problems. The system can also discover the risks of combined medication that are difficult to detect by relying solely on the human brain through correlation analysis of cross-domain rules. Unlike traditional "post-event quality control", this intelligent audit mechanism is a pre-event warning, just like setting up a "firewall" for doctors' decision-making, avoiding the wrong prescribing and execution of defective prescriptions from the source, and maximizing the safety of TPN treatment. It is worth mentioning that the system's knowledge graph is not static, but is dynamically updated through real-time crawling of literature and drug instructions. Once drug specifications and medical standards change, the knowledge graph will also evolve, triggering the automatic upgrade of audit rules to ensure that they keep pace with the times. Using NLP and causal reasoning technology, the system can also continuously learn from adverse event reports, explore the root causes of errors, and form more comprehensive and fine-grained risk rules, so that the coverage of intelligent warnings continues to expand and the accuracy continues to improve.

[0064] 3. Intelligent preparation and quality control based on the Internet of Things: The preparation of TPN liquid medicine is a very complex and delicate process, involving the weighing, dissolution, mixing, sterilization, and packaging of various drugs. Each link must be intelligently controlled and monitored in real time to ensure the safety and effectiveness of the final product. This system fully draws on the concept of industrial automated production lines and uses the Internet of Things architecture of "sensor + automation device + real-time monitoring system" to realize the digital control of the entire process of drug preparation. On the automated preparation platform, the system first uses RFID electronic tags and smart scales to accurately control the feeding amount of various raw materials, and realizes the matching verification of raw materials and prescriptions through barcode scanning. During the preparation of the liquid medicine, the system collects key process parameters such as temperature, pH value, turbidity, and particle count in real time, and uploads them to the intelligent quality control system in real time through wireless transmission. According to the process parameter specifications, the quality control system uses statistical process control and other technologies to automatically analyze and warn of abnormalities for various indicators. Once the risk of exceeding the standard is found, the process engineer is notified to suspend production and isolate the problem product, thereby preventing unqualified products from flowing to patients. After the preparation is completed, each bag of liquid medicine is automatically affixed with a unique electronic label containing the batch, composition, preparation time, expiration date, execution link flow information, etc. of the liquid medicine, realizing the refined management of "one liquid, one code". The subsequent storage, transportation, use and other links strictly implement the "scan code verification" procedures to ensure that each bag of liquid medicine is circulated under a controlled state. This intelligent preparation and quality control solution that applies the concepts of the Internet of Things and industrial automation not only reduces the error rate of manual operation, but also realizes the visualization and traceability of the entire preparation process through data management, forming a closed-loop, self-checking quality control system, which significantly improves the uniformity and reliability of the quality of TPN liquid medicine.

[0065] 4. Intelligent infusion management based on closed-loop control: In clinical practice, the effect and safety of intravenous infusion depends not only on the drug solution itself, but also on many operating parameters such as infusion speed, sequence, and duration. However, most of the current infusion pumps can only perform simple fixed-speed infusion tasks and do not have the ability to implement dynamic adjustments based on the patient's real-time physiological changes. This system innovatively integrates physiological feedback control and intelligent algorithms into infusion management, and constructs a closed-loop control mechanism of "monitoring-analysis-regulation-monitoring". The smart wristband worn by the patient collects his or her vital signs data such as heart rate, blood pressure, and blood oxygen in real time, and measures the dynamic changes of venous pressure and blood viscosity through micropumps and centrifugal sensors. These signals are transmitted to the intelligent infusion system in real time, integrated with the patient's pharmacokinetic model, and evaluate the body's tolerance to the drug solution. The system designs a set of intelligent speed regulation algorithms based on fuzzy control, which comprehensively considers the above parameters, dynamically calculates the optimal infusion speed, and automatically adjusts the speed of the infusion pump. The whole process does not require human intervention, and the control strategy strictly follows the embedded safety rules. Once the patient's adverse reaction is detected, the system can make an autonomous decision to stop the infusion and send an alarm to the doctor on duty to minimize the occurrence of serious complications. It should be emphasized that the closed-loop control system is not a simple linear feedback, but a continuous optimization of the control strategy through reinforcement learning. According to the patient's dynamic response to treatment, the system evaluates the effect of the current control strategy through the objective function (such as the stability of physiological indicators), and uses the reinforcement learning algorithm to seek a balance between "exploration" and "utilization". Under the premise of ensuring safety, it continuously tries to optimize the combination of infusion modes, find the "sweet spot" of treatment effect and tolerance, and form the best personalized infusion strategy. It can be said that this intelligent infusion system allows TPN treatment to be truly "tailored" and keeps the treatment process in a dynamic optimal state through real-time regulation.

[0066] 5. Blockchain-based full-process data closed-loop management: TPN treatment involves many links such as medical orders, solution preparation, infusion, and monitoring. Each link will generate a large amount of heterogeneous data. How to ensure the integrity, consistency, and security of these data in the whole process is a major problem. Traditional centralized information systems have inherent defects such as data islands and privacy leaks, making it difficult to achieve trusted full-process closed-loop management of data. This system innovatively introduces blockchain distributed ledger technology, hashes the original data of each link of TPN treatment on the chain, and builds a set of tamper-proof and traceable data chains based on cryptographic principles. Each newly generated data block is connected to the previous block through a hash, and any change in the data will change the fingerprint map of the entire chain, thereby achieving tamper-proof. At the same time, the data is distributed and stored in the chain in ciphertext form. Only authorized subjects can decrypt and access it, and all access records are automatically uploaded to the chain, ensuring the security and privacy of the data throughout its life cycle. The smart contract module in the blockchain can automatically manage data access strategies and strictly control the operating permissions of each participant. For example, doctors can only issue medical orders, pharmacists can only prepare drugs, and nurses can only perform infusions. Each operation requires consensus from multiple parties to take effect, avoiding unauthorized operations. At the same time, smart contracts can also perform logical verification of cross-link data to ensure the consistency of data between links. Once the data in the links of medical orders, liquid preparation, infusion, etc. are found to be mismatched, the system will immediately issue an automatic warning, triggering manual review and error analysis. Thanks to the trustworthy, tamper-proof and traceable characteristics of blockchain, this system has built a set of self-verifying and non-repudiable data closed loops, realizing a trusted evidence chain of multi-party collaboration among patients, doctors, pharmacists, and nurses, which not only facilitates the responsibility tracing of medical disputes, but also provides a data basis for continuous quality improvement. It can be foreseen that blockchain will become a new paradigm for smart medical data management.

[0067] 6. Intelligent decision optimization based on causal reasoning: At present, clinical pathways, medical standards and expert consensus are mostly presented in a simple "if-then" form, lacking in-depth explanation of the causal relationship between treatment methods and disease prognosis, and it is difficult to directly guide dynamic and personalized clinical decision-making. This system uses causal reasoning technology to conduct association analysis and causal modeling on massive data in various links of TPN treatment, explore the internal mechanism hidden behind the data, form a multi-level causal decision network between "drug composition-dose-infusion mode-disease outcome", and draw an intuitive causal decision diagram to provide a reference for formulating the optimal treatment strategy. For example, the system found that some patients with low albumin and hypokalemia deteriorated after large-dose, rapid infusion of amino acids. Through medical causal analysis, it was found that it might be related to metabolic acidosis, so it was recommended to adopt a small-dose, slow infusion strategy for patients with similar characteristics, which significantly improved the prognosis. For example, for patients with renal insufficiency, the system concluded through causal reasoning that the strategy of "limiting total fluid volume-adjusting electrolyte concentration-monitoring urine volume" is better than the conventional TPN scheme, and quantitatively analyzed the probability distribution of recovery under different dose and speed combinations, thereby forming the best treatment curve. More importantly, the system does not regard this strategy based on causal reasoning as a static "gold standard", but continuously verifies and corrects its effectiveness. On the one hand, the system associates the treatment parameters and outcome data of each patient in real time, and dynamically updates the probability of existing causal relationships through Bayesian inference. On the other hand, the system continuously learns medical literature and guidelines, extracts new causal evidence and optimizes the existing decision map. As a result, the system has established an open and evolving causal knowledge base, which can provide personalized and evidence-based suggestions for TPN decisions under various disease states, and enable "empirical medicine" to transition to "evidence-based medicine". From "black box" to "white box", from "correlation" to "causality", this causal reasoning-driven intelligent decision optimization mechanism will surely lead to a paradigm revolution in clinical decision support systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a framework diagram of the total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technologies of the present invention. DETAILED DESCRIPTION

[0069] The following will combine the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all the embodiments. All other embodiments obtained by ordinary technicians in this field without creative work based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0070] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "top / bottom" indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0071] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "provided with", "mounted / connected", and "connected" should be understood in a broad sense. For example, "connected" means fixed connection, detachable connection, or integral connection; mechanical connection, electrical connection; direct connection, indirect connection through an intermediate medium, and internal communication between two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention are understood according to specific circumstances.

[0072] Example 1

[0073] Please refer to Figure 1 , a total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology, including an intelligent TPN formula module 101, a pharmaceutical intelligent review module 102, an intelligent preparation and quality control module 103, an intelligent infusion management module 104, a real-time dynamic monitoring module 105 and a full-process closed-loop management module 106;

[0074] The intelligent TPN prescription module 101 is used to intelligently generate personalized TPN prescriptions based on the patient's clinical data;

[0075] The pharmaceutical intelligent review module 102 is used to review the rationality of TPN prescriptions, identify inappropriate prescriptions, and give revision suggestions;

[0076] The intelligent preparation and quality control module 103 is used to automatically generate preparation instructions according to the reviewed TPN prescription and perform real-time quality monitoring on the preparation process;

[0077] The intelligent infusion management module 104 performs precise control and real-time safety monitoring of the TPN infusion process to ensure the accuracy and safety of the infusion process;

[0078] The real-time dynamic monitoring module 105 continuously monitors the patient's vital signs, biochemical indicators, and metabolic status in real time, and dynamically adjusts the TPN treatment plan accordingly;

[0079] The full-process closed-loop management module 106 is used for data collection, storage, sharing and tracing of the entire TPN treatment process, forming a data closed loop and promoting continuous optimization.

[0080] This intelligent TPN full-process control system fully applies artificial intelligence technology, with "closed-loop quality control" and "continuous optimization" as the core concepts, seamlessly connecting key links such as clinical decision-making, prescription review, drug preparation, infusion configuration, infusion management, and nutritional monitoring to form a highly integrated, intelligent and collaborative integrated solution. The modules in the system coordinate with each other, and strict logical relationships are set in key links to ensure the continuity of data flow and the consistency of prescription information. At the same time, the system makes full use of big data analysis and machine learning technology to continuously learn and optimize in medical data, and continuously improve the accuracy and safety of TPN treatment. We believe that the establishment and application of this system will significantly improve the hospital's TPN diagnosis and treatment level, reduce medical errors, improve patient prognosis, and be at the forefront of smart medical care. This will surely create a new model of clinical nutritional treatment and provide valuable experience for the research and development of intelligent auxiliary medical systems.

[0081] Through the above detailed definition of the functions of each module and the description of the implementation process, it can be seen that the intelligent TPN control system is a highly integrated and closely coordinated whole. From prescription optimization to pharmaceutical review, from preparation quality control to infusion management, from dynamic monitoring to closed-loop management, each link is seamlessly connected and linked together. The data flow is transferred between the modules, driving the full process of intelligent closed-loop control of TPN treatment. The modules are not simply linearly connected in series, but form a causal and spiral feedback control loop. The intelligent formula is continuously iterated and optimized based on the results of pharmaceutical review, and the pharmaceutical review continuously improves the rule base based on this; dynamic monitoring triggers the real-time adjustment of infusion control, and the massive data of the infusion process feeds back to the continuous learning of the monitoring model; closed-loop management promotes the continuous improvement of each link through full-process data analysis, and the improvement results of each link are reflected in the data, forming a virtuous circle. It is this synergy among modules that makes the intelligent TPN management and control system an organic whole that evolves and continuously optimizes itself. It can adapt to the dynamic changes of patients, meet their personalized needs, ensure the safety of their medication, improve the overall diagnosis and treatment efficiency and quality, and ultimately achieve precise and intelligent TPN treatment.

[0082] The intelligent TPN formula module 101 first automatically extracts the patient's demographic characteristics, body composition, biochemical indicators, and diagnostic data through the electronic medical record system, and standardizes them into a format that the system can recognize; then, the pre-built nutritional risk assessment model is used to score the patient's nutritional risk, referred to as NRS, and the NRS is used as an important reference for subsequent formula optimization; then, the standardized patient data is input into the formula optimization model, and the pharmaceutical properties of the TPN components, such as calories, protein, and electrolyte content parameters, are retrieved from the drug information database, and the sequential quadratic programming algorithm is used to solve the problem, and an optimal TPN formula prescription that takes into account both patient needs and pharmaceutical requirements is obtained. Finally, the generated TPN prescription is sent to the pharmaceutical intelligent review module 102 for rationality review. The data flow is as follows: EMR->patient data extraction and standardization->nutritional risk assessment->formula optimization solution->TPN prescription->pharmacy intelligent review.

[0083] The pharmacy intelligent review module 102 receives the TPN prescription from the intelligent TPN formula module 101, converts the prescription into a structured form using natural language processing technology, and then retrieves the drug interaction knowledge base and the incompatibility library, and uses a rule-based reasoning algorithm to review the prescription, focusing on the following issues: whether the prescription components are consistent with the patient's condition, whether the drug dosage is appropriate, drug interactions, and incompatibility. For prescriptions that pass the review, they are directly sent to the intelligent preparation and quality control module 103; for prescriptions that fail the review, the system will mark specific problems and give revision suggestions, which will be fed back to the upstream intelligent TPN formula module 101 for optimization and iteration until a qualified TPN prescription is generated. The data flow is TPN prescription->prescription structuring->rationality review->pass->intelligent preparation and quality control; fail->problem marking and revision suggestions->intelligent TPN formula optimization and iteration.

[0084] The intelligent preparation and quality control module 103 receives the qualified TPN prescription from the pharmaceutical intelligent review module 102, automatically generates a standardized preparation sheet, and transmits it to the automated preparation equipment; at the same time, according to the prescription requirements, it sends a drug collection instruction to the drug warehouse management system, and the drug warehouse management system feeds back the collected drug information to the intelligent preparation submodule for distribution and preparation; when the automated preparation equipment performs TPN configuration, various sensors such as mass flow meters, pH meters, turbidity meters, particle detectors, etc. are used to collect configuration process parameters in real time, and compare them with the preset quality standard thresholds, and monitor key quality attributes such as drug solution ratio accuracy, drug solution clarity, and particle quantity in real time. Once an abnormal deviation is found , the system immediately issues an early warning, and automatically isolates the problematic preparations, and at the same time feeds back the abnormal information to the pharmaceutical intelligent audit module 102, triggering the re-examination and optimization of the TPN prescription; for the qualified TPN preparations, the label is automatically printed, and the label contains patient information, drug information, preparation time, validity period, storage conditions, preparer, and reviewer information, and is presented in the form of a barcode, and the qualified finished liquid medicine is loaded into the intelligent infusion vehicle, and the corresponding relationship information between the liquid medicine and the patient and the infusion pump is written into the control panel of the infusion vehicle, and the infusion vehicle is transported to the ward, and the bagging completion signal is transmitted to the intelligent infusion management module 104, and the preparation process data is uploaded to the full-process closed-loop management module 106 for storage and analysis. The data flow is TPN prescription->dispensing order generation->pharmacy warehouse drug collection->drug distribution->automatic preparation->quality monitoring data collection->real-time quality assessment->qualified->bagging and labeling->infusion management; unqualified->early warning and isolation->pharmacy audit trigger->prescription re-optimization.

[0085] The intelligent infusion management module 104 receives the prepared TPN solution from the intelligent preparation and quality control module 103. When the nurse inserts the infusion bag containing the patient's TPN solution into the infusion pump in the ward, the infusion pump automatically reads the solution information and matches it with the doctor's order by scanning the barcode on the infusion bag. If the match is successful, the infusion pump automatically extracts the infusion parameters, such as infusion speed, infusion time, and infusion volume, from the doctor's order after intelligent review, and automatically sets them. The nurse starts the infusion after checking again. During the infusion process, the medical staff monitors the various sensors connected to the patient's venous catheter, such as pressure sensors, bubble sensors, and blood oxygen saturation sensors, and the relevant monitoring data is automatically uploaded to the intelligent infusion management module 104. If the infusion line is detected to be detached, blocked, or leaking, an alarm is issued, and corresponding measures are taken intelligently according to the abnormal type, such as automatically pausing the infusion, adjusting the infusion speed, and notifying the nurse on duty to go and check. When the nurse is patrolling, the barcode of the infusion bag is scanned by the handheld terminal to obtain the current infusion progress and remaining time information. The results of each inspection are automatically recorded in the intelligent infusion management module 104; after the infusion is completed, the infusion pump is automatically stopped and the infusion record is saved, including the drug name, infusion time, infusion volume, average infusion speed, and infusion process curve, for clinical query and medical accident tracing. All infusion process data will be uploaded to the full-process closed-loop management module 106 in real time for audit tracing. The data flow is TPN preparation -> infusion parameter setting -> infusion start -> real-time collection of vital signs and infusion parameters -> real-time dynamic monitoring and analysis -> normal -> maintenance infusion -> infusion end -> infusion process data -> full-process closed-loop management; abnormal -> alarm -> automatic adjustment or stop infusion -> abnormal data -> full-process closed-loop management.

[0086] The real-time dynamic monitoring module 105 continuously collects the patient's heart rate, blood pressure vital signs data, as well as blood sugar, electrolytes, and acid-base balance biochemical indicators through the interface with the ward monitoring equipment and the inspection system; these data will be transmitted to the intelligent infusion management module 104 in real time for evaluating the safety of the infusion process; the real-time dynamic monitoring module 105 presents the patient's vital signs trend data, such as heart rate, blood pressure, etc., and laboratory test results, such as blood routine, urine routine, electrolytes, coagulation function, liver and kidney function, etc., in a rasterized manner, so that medical staff can keep abreast of the patient's condition at any time; when vital signs, such as heart rate and blood pressure, show abnormal fluctuations, or When the patient's laboratory test results, such as white blood cell count, blood sugar, blood potassium concentration, etc., exceed the preset threshold, the system pushes an alarm message to the doctor on duty and gives a reference diagnosis and treatment suggestions; at the same time, the module uses a machine learning algorithm to continuously calculate the patient's comprehensive nutritional score nPNI in the background to evaluate the patient's overall nutritional status; when nPNI is lower than the normal value, the system infers that the patient may have hypoproteinemia, anemia, inflammatory response and nutrition-related problems, and adjusts the TPN prescription accordingly; the adjusted TPN prescription re-enters the review process of the intelligent TPN formula module 101 until an optimized new prescription is generated, and the nutritional status evaluation adopts the following formula:

[0087] nPNI(t)=f(HGB(t), PALB(t), PCR(t),...)

[0088] nPNI(t) represents the nutritional status assessment at time t, taking into account multiple nutrition-related parameters such as hemoglobin HGB(t), serum albumin PALB(t), and protein degradation rate PCR(t);

[0089] The prescription adjustment adopts the following formula: If nPNI(t)<nPNIl,

[0090] Then TPN(t+1)=a(t)·Pr+b(t)·Ft+c(t)·Cb

[0091] When nPNI(t) is lower than the threshold nPNIl, the nutritional prescription TPN(t+1) for the next day (t+1) needs to be adjusted according to the current nPNI and the TPN(t) of the previous day; Pr, Ft, Cb here represent the amount of protein, fat and carbohydrates respectively; coefficients a(t), b(t), c(t) are weight coefficients dynamically fitted according to nPNI(t) and TPN(t), which are used to adjust the proportion of various nutrients. The data flow is vital signs data->infusion safety assessment->intelligent infusion management; biochemical indicators->metabolic state analysis->TPN plan re-evaluation->dynamic adjustment suggestions->doctor review->new prescription->intelligent TPN formula; abnormal warning->doctor decision-making->emergency measures.

[0092] The full-process closed-loop management module 106 collects all data generated in each link of TPN treatment through the hospital-wide information integration platform, including prescription data, dispensing data, infusion data, and monitoring data, and uses blockchain technology to establish a TPN full-process data chain to ensure the integrity, accuracy, immutability and traceability of the data; the full-process closed-loop management module 106 implements TPN full-process closed-loop management based on the five basic principles of 5R, namely Right patient, Right drug, Right dose, Right route, and Right time; the full-process closed-loop management module 106 intuitively displays the patient's complete TPN treatment history in the form of a timeline, spanning multiple links such as outpatient, hospitalization, and discharge; each node of the timeline can be expanded to present key information, such as the time of doctor's prescription, prescription composition, configuration and adjustment records, drug traceability codes, infusion curves, and nutritional monitoring reports, for easy review of the entire process; each link strictly follows the "SCANNING IN, SCANNING "OUT", referred to as SISO specification, that is: scan the patient's wristband at the beginning of each key link, confirm the patient's identity before executing the doctor's order; at the end of each key link, scan the patient's wristband and the doctor's order again, and submit the execution result to form a closed loop; require two people to sign the key links of doctor's order, drug preparation, infusion configuration, and infusion pipeline verification, and the second signer must be authenticated by fingerprint recognition or iris recognition to eliminate the phenomenon of "signing on behalf of others"; each step of the operation forms a traceable electronic signature; all original documents, such as doctor's order, preparation, and infusion orders, are printed on special anti-tampering paper and affixed with a "one-time seal" with a unique traceability code printed on the seal; each document must be fully archived after execution for later reference; the embedded TPN quality continuous improvement submodule can implement T2-1 statistical process control, i.e. SPC, for the key indicators of doctor's order standardization, preparation accuracy, and infusion error rate. According to the SPC control chart intelligent early warning and continuous optimization, all data are stored in the central database in a unified format and standard, and shared according to the predetermined data access control strategy for use by relevant departments and management departments. The pharmacy department can access prescription and dispensing data to evaluate the rationality of prescriptions and the quality of dispensing; the nursing department can access infusion and monitoring data to evaluate the standardization and safety of infusion management; the medical management department can access the full process data to evaluate the overall diagnosis and treatment quality and efficiency. The full process closed-loop management module 106 is also embedded with data analysis and mining tools to regularly analyze the full process data, generate quality reports, identify potential problems, and make continuous improvement suggestions, such as analyzing the medical prescription medication pattern, discovering unreasonable prescription patterns and standardizing them; analyzing the clinical efficacy of different prescriptions and optimizing the TPN prescription library; analyzing the relationship between different infusion strategies and the incidence of adverse events, and formulating best infusion practice specifications.The data flow is prescription data -> dispensing data -> infusion data -> monitoring data -> blockchain data uploading -> central database storage -> data sharing -> data application in various departments -> data analysis and mining -> quality reporting and continuous improvement. Use blockchain technology to build a TPN full-process data chain:.

[0093] Block i ={Hash i-1 ,Data i ,Time i ,Hash i}

[0094] Among them, Hash i is the hash value of block i, Data i is the data of block i, Time i It is a timestamp. Blockchain ensures the immutability and traceability of data, and realizes closed-loop management of data throughout the entire process.

[0095] Input age A, gender G, height H, weight W; body composition: muscle mass M, fat mass F, water content H2O; biochemical indicators: albumin ALB, lymphocyte count LY, blood sugar GLU, renal function eGFR, liver function ALTAST; clinical diagnosis: primary disease D1, complications D2, surgical history SH, infection status INF;

[0096] The nutritional risk assessment model uses principal component analysis and logistic regression methods:

[0097] NRS=f(A,G,H,W,M,F,H2O,ALB,LY)=β 0 +β 1 PC 1 +β 2 PC 2 +...+β n PC r

[0098] Among them, PC i is the i-th principal component, β i is the corresponding regression coefficient; the objective function is to minimize NRS while meeting nutrient requirements:

[0099] Min INRS

[0100] st

[0101] Calorie requirement: HC (kcal / d) = BEE × AF × IF × TF

[0102] Protein requirement: PR (g / kg / d) = IBW × (1.2-2.0)

[0103] Fat energy supply ratio: 20%~30%

[0104] Carbohydrate energy supply ratio: 50% to 70%

[0105] Electrolyte concentration limit: Na + (135-145mmol / L); K + (3.5-5.5mmol / L)

[0106] Maximum infusion volume: 40mL / kg / d

[0107] Osmotic pressure limit: 900mOsm / L

[0108] Drug incompatibility: Incompatible drug pairs {(D1,D2),(D3,D4),...}

[0109] The sequential quadratic programming algorithm is used to solve the optimal TPN formula:

[0110] TP={V,AA,LIP,GLU,Na + ,K + ,...}

[0111] Among them, V is the TPN infusion volume, AA is the amino acid volume, LIP is the fat emulsion volume, and GLU is the glucose volume. SQP first solves the quadratic approximation of the objective function, then solves the linear approximation of the constraints, iterates until the conditions converge, and outputs the dosage of each nutrient component of the optimal TPN formula.

[0112] The medical order structuring adopts a natural language processing algorithm, and the free text of the TPN medical order is input. The specific steps are as follows:

[0113] (1) Word segmentation and part-of-speech tagging: Segment the medical order text into words and mark the part of speech of each word;

[0114] (2) Named entity recognition: identifying key entities such as drug name, dosage number, unit, and route of administration;

[0115] (3) Relationship extraction: extracting key attribute pairs of drug-dose and drug-administration route;

[0116] (4) Attribute normalization: Map synonyms, abbreviations, and typos to a standard vocabulary;

[0117] (5) Five-tuple filling: fill in the standard template of drug, dosage, route of administration, frequency, and course of treatment according to the extracted attribute pairs, and mark the attributes that cannot be filled as null values;

[0118] Output a structured quintuple consisting of drug, dose, route of administration, frequency, and duration of treatment.

[0119] The rule-based reasoning includes the following steps:

[0120] Knowledge base: (1) Drug interaction knowledge base: define groups of if-then rules, such as: If {diazepam, midazolam}∈prescription Then alarm "diazepam and midazolam should not be used at the same time";

[0121] (2) Incompatibility database: defined as if-then rules; for example, If {amino acid, fat emulsion}∈prescription & concentration (amino acid)>6% Then alarm "It is not suitable to use fat emulsion when the amino acid concentration is too high";

[0122] (3) Dose range library: defined as a value interval, for example, the glucose concentration for children should be between 5% and 12%;

[0123] Reasoning process: (1) Matching mode: match the structured medical instructions with the rule conditions of the knowledge base; (2) Conflict detection: check whether there is a logical conflict in the conclusion part of the rule; (3) Problem output: output the problem description and revision suggestions of the hit rule.

[0124] We designed a clinical control trial to verify the clinical superiority of this intelligent TPN management and control system by comparing the implementation example with the control example.

[0125] 1. Experimental Design

[0126] 1. Subjects: Inclusion criteria: patients undergoing elective gastrointestinal tumor surgery, aged 18-70 years, ECOG score ≤ 2 points, requiring TPN support ≥ 7 days. Exclusion criteria: patients undergoing transplant surgery, patients with concurrent infection, and patients with severe organ dysfunction.

[0127] 2. Grouping scheme: Implementation group: adopt this intelligent TPN management and control system. Control group 1: adopt traditional TPN empirical prescription. Control group 2: adopt the goal-oriented TPN scheme guided by intravenous nutrition risk screening (NRS2002). Control group 3: adopt the scheme recommended by the American Society for Parenteral and Enteral Nutrition (ASPEN) guidelines.

[0128] 3. Sample size: Based on previous studies and preliminary experimental results, with an effect size of increasing the effective rate by 10%, α=0.05, β=0.10, and an optimal design, the sample size of each group was calculated to be 50 cases, for a total of 200 cases.

[0129] 2. Observation indicators and detection methods

[0130] 1. Nutritional status: (1) Body mass index (BMI): weight (kg) / height (m) 2 , tested at enrollment and on days 7 and 14. (2) Prealbumin (PAB): turbidimetric method, tested at enrollment and on days 3, 7, 10, and 14.

[0131] 2. Incidence of metabolic complications: (1) Hyperglycemia: fasting blood glucose > 6.1mmol / L or random blood glucose > 11.1mmol / L. (2) Hyperkalemia: serum potassium > 5.5mmol / L. (3) Hypophosphatemia: serum phosphorus < 0.8mmol / L. (4) Metabolic acidosis: pH < 7.35 and blood HCO 3 -<22mmol / L. The above indicators are tested daily.

[0132] 3. Incidence of infectious complications: (1) Catheter-related bloodstream infection (CRBSI): semi-quantitative method. (2) Pneumonia: clinical diagnosis + imaging diagnosis. (3) Incisional infection: positive bacterial culture of incisional secretions. Infection-related indicators were assessed daily, and corresponding examinations were performed when infection was suspected.

[0133] 4. Hospitalization time: the number of natural days from admission to discharge. 5. Hospitalization expenses: the total hospitalization expenses recorded on the first page of the medical record.

[0134] 3. Results

[0135]

[0136]

[0137] Table 1

[0138] Note: Data are expressed as mean ± standard deviation or percentage. Inter-group comparison was performed using analysis of variance or chi-square test. P < 0.05 was considered statistically significant.

[0139] As can be seen from the table, the embodiment group was significantly superior to the control group (P<0.05) in improving the nutritional status of patients (increasing BMI, PAB levels), reducing the incidence of metabolic complications (hyperglycemia, hyperkalemia, hypophosphatemia, metabolic acidosis), controlling the risk of infectious complications (CRBSI, pneumonia, wound infection), shortening the length of hospital stay, and saving hospitalization costs, which fully confirmed the significant clinical advantages of this intelligent TPN management and control system.

[0140] IV. Results Analysis

[0141] As shown in Table 1, the control group 1 generally had a high malnutrition rate at enrollment and no obvious improvement in nutritional status during hospitalization, indicating that the empirical TPN prescription was not effective in optimizing the nutritional status of patients. The malnutrition rate of the embodiment group was low at enrollment, and the nutritional status of patients continued to improve with TPN support, indicating that the intelligent TPN management and control system can more accurately assess the nutritional risk of patients and formulate individualized TPN plans to prevent and correct malnutrition.

[0142] Although the control groups 2 and 3 used NRS2002 and ASPEN guidelines to guide TPN, their metabolic complications, especially the incidence of hyperglycemia, were still significantly higher than those in the embodiment group, suggesting that existing nutrition screening tools and clinical guidelines are still insufficient in dynamically managing the risk of TPN complications. The dynamic infusion optimization strategy implemented by this system based on machine learning and physiological feedback control shows unique advantages in reducing TPN metabolic complications.

[0143] The incidence of catheter-related bloodstream infection in control group 1 was significantly higher than that in other groups, indicating that blind, empirical TPN may bring serious infection risks. Although preventive guidelines were introduced in control groups 2 and 3, the infection incidence was still higher than that in the implementation group, indicating that TPN without intelligent quality control and blockchain management cannot fundamentally avoid infection risks. This system standardizes TPN operations throughout the entire process, making its risks controllable, its sources traceable, and its responsibilities traceable, greatly improving the safety of infusion therapy.

[0144] The hospital stay of the embodiment group was significantly shorter than that of the control group, and the hospitalization cost was relatively low. This is due to the fact that the system effectively reduced various complications, improved patient prognosis, and improved medical efficiency through full-process closed-loop intelligent management. From the perspective of the hospital, this system can also save medical resources and create more social benefits.

[0145] V. Best practice: Further analysis of the practice group data revealed that the following combination of factors / measures was associated with the best clinical benefit: 1) Complete nutritional risk screening (NRS) within 24 hours of admission and formulate the first TPN prescription accordingly; 2) Use a three-chamber bag solution with 50 g / d protein, 50:50 sugar-to-lipid ratio, and ≤6% amino acids; 3) Based on dynamic blood glucose monitoring, use fuzzy control algorithm to achieve accurate insulin titration; 4) Rescreen NRS scores every 2-3 days and dynamically adjust TPN prescriptions; 5) Three-chamber Bags with trace elements (zinc, copper, manganese, chromium) and vitamins (VitB, VitC); 6) Standardized drug selection for TPN (avoid high-sugar and hyperosmotic preparations), and set the maximum concentration warning value; 7) Aseptic operation throughout the process, and daily assessment of the catheter; 8) Use an intelligent infusion pump + automatic reset device, and set a pressure threshold alarm; 9) Review liver and kidney function on the third day of treatment, and adjust the use of fat emulsion; 10) Provide TPN home care and follow-up guidance upon discharge, and inform the key points of complication prevention.

[0146] VI. Conclusion Through rigorous clinical controlled trials, the clinical superiority of this intelligent TPN management and control system has been fully verified. Through a series of intelligent innovative measures such as intelligent assessment of nutritional risks, personalized optimization of formulas, quality control of the preparation process, and dynamic feedback of infusion strategies, it has shown significant effects in improving patients' nutritional status, reducing metabolic complications, controlling infection risks, shortening hospital stays, and saving medical costs. It provides a new solution for improving the level of clinical nutritional treatment and ensuring the safety of patients' medication. In the future, this system is expected to further expand its application scenarios and become a benchmark case for smart healthcare.

[0147] The above description is only a preferred specific implementation manner of the present invention; however, the protection scope of the present invention is not limited thereto; any technician familiar with the technical field within the technical scope disclosed by the present invention; any replacement or change made by the technical solution and improved concept of the present invention shall be covered by the protection scope of the present invention.

Claims

1. A total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology, characterized by: It includes intelligent TPN formula module, intelligent pharmaceutical audit module, intelligent preparation and quality control module, intelligent infusion management module, real-time dynamic monitoring module and full-process closed-loop management module; The intelligent TPN prescription module is used to intelligently generate personalized TPN prescriptions based on the patient's clinical data; The pharmaceutical intelligent review module is used to review the rationality of TPN prescriptions, identify inappropriate prescriptions, and provide revision suggestions; The intelligent preparation and quality control module is used to automatically generate preparation instructions based on the reviewed TPN prescription and perform real-time quality monitoring of the preparation process; The intelligent infusion management module performs precise control and real-time safety monitoring of the TPN infusion process to ensure the accuracy and safety of the infusion process; The real-time dynamic monitoring module continuously monitors the patient's vital signs, biochemical indicators, and metabolic status in real time, and dynamically adjusts the TPN treatment plan accordingly; The full-process closed-loop management module is used for data collection, storage, sharing and tracing of the entire TPN treatment process, forming a data closed loop and promoting continuous optimization; The intelligent preparation and quality control module receives the qualified TPN prescription from the pharmaceutical intelligent review module, automatically generates a standardized preparation sheet, and transmits it to the automated preparation equipment; at the same time, according to the prescription requirements, it sends a drug collection instruction to the drug warehouse management system, and the drug warehouse management system feeds back the collected drug information to the intelligent preparation submodule for distribution and preparation; when the automated preparation equipment performs TPN configuration, various sensors are used to collect configuration process parameters in real time, and compare them with the preset quality standard thresholds, and the solution ratio, particle number, and stability key quality attributes are monitored in real time. Once an abnormal deviation is found, the system immediately issues an early warning and automatically Isolate problematic preparations and feed back abnormal information to the pharmaceutical intelligent audit module, triggering a re-examination and optimization of TPN prescriptions; automatically print labels for qualified TPN preparations, which contain patient information, drug information, preparation time, expiration date, storage conditions, preparer, and reviewer information, and are presented in the form of barcodes; load qualified finished drug solutions into smart infusion carts, write the corresponding relationship information between drug solutions and patients and infusion pumps into the control panel of the infusion carts, transport the infusion carts to the wards, and transmit a bagging completion signal to the intelligent infusion management module; upload the preparation process data to the full-process closed-loop management module for storage and analysis.

2. The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology according to claim 1, characterized in that: The intelligent TPN formula module first automatically extracts the patient's demographic characteristics, body composition, biochemical indicators, and diagnostic data through the electronic medical record system, and standardizes them into a format recognizable by the system; then, a pre-constructed nutritional risk assessment model is used to perform a nutritional risk score on the patient, referred to as NRS, and the NRS is used as an important reference for subsequent formula optimization; then, the standardized patient data is input into the formula optimization model, and the pharmaceutical properties of the TPN components are retrieved from the drug information database. The pharmaceutical properties include calorie, protein, and electrolyte content parameters, and the sequential quadratic programming algorithm is used to solve them to obtain an optimal TPN formula prescription that takes into account both patient needs and pharmaceutical requirements. Finally, the generated TPN prescription is sent to the pharmaceutical intelligent review module for rationality review.

3. The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology according to claim 1 is characterized in that: The pharmaceutical intelligent review module receives the TPN prescription from the intelligent TPN formula module, converts the prescription into a structured form using natural language processing technology, then retrieves the drug interaction knowledge base and the incompatibility library, and uses a rule-based reasoning algorithm to review the prescription, focusing on the following issues: whether the prescription components are consistent with the patient's condition, whether the drug dosage is appropriate, drug interactions, and incompatibility. For prescriptions that pass the review, they are directly sent to the intelligent dispensing and quality control module; for prescriptions that fail the review, the system will mark the specific problems and give revision suggestions, which will be fed back to the upstream intelligent TPN formula module for optimization and iteration until a qualified TPN prescription is generated.

4. The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology according to claim 1, characterized in that: The intelligent infusion management module receives the prepared TPN solution from the intelligent preparation and quality control module. When the nurse inserts the infusion bag containing the patient's TPN solution into the infusion pump in the ward, the infusion pump automatically reads the solution information and matches it with the doctor's order by scanning the barcode on the infusion bag; if the match is successful, the infusion pump automatically extracts the infusion parameters from the doctor's order after intelligent review, where the infusion parameters include infusion speed, infusion time, and infusion volume, and automatically sets them; the nurse starts the infusion after checking again; during the infusion process, the medical staff monitors the patient's venous catheter through various sensors, including pressure sensors, bubble sensors, and blood oxygen saturation sensors, and the relevant monitoring data is automatically uploaded to the intelligent Infusion management module; if abnormal conditions such as detachment, blockage, and leakage of the infusion line are detected, an alarm will be issued, and corresponding measures will be taken intelligently according to the type of abnormality, including automatic suspension of infusion, adjustment of infusion speed, and notification of the nurse on duty to check; when the nurse makes an inspection, she scans the barcode of the infusion bag through a handheld terminal to obtain the current infusion progress and remaining time information; each inspection result is automatically recorded in the intelligent infusion management module; after the infusion is completed, the infusion pump will automatically stop and save the infusion record, including the drug name, infusion time, infusion volume, average infusion speed, and infusion process curve, for clinical inquiry and medical accident tracing, and all infusion process data will be uploaded to the full-process closed-loop management module in real time for audit tracing.

5. The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology according to claim 1 is characterized in that: The real-time dynamic monitoring module continuously collects the patient's heart rate, blood pressure vital signs data, as well as blood sugar, electrolytes, and acid-base balance biochemical indicators through the interface with the ward monitoring equipment and the inspection system; these data will be transmitted to the intelligent infusion management module in real time to evaluate the safety of the infusion process; the real-time dynamic monitoring module presents the patient's vital signs trend data and laboratory test results in a rasterized manner, so that medical staff can grasp the patient's condition at any time; when the vital signs fluctuate abnormally, or the laboratory test results exceed the preset threshold, the system pushes an alarm message to the doctor on duty and gives a reference diagnosis and treatment suggestions; at the same time, the module uses a machine learning algorithm to continuously calculate the patient's comprehensive nutritional score nPNI in the background to evaluate the patient's overall nutritional status; when nPNI is lower than the normal value, the system infers that the patient may have hypoproteinemia, anemia, inflammatory response nutrition-related problems, and adjusts the TPN prescription accordingly; the adjusted TPN prescription re-enters the review process of the intelligent TPN formula module until an optimized new prescription is generated, and the nutritional status evaluation adopts the following formula: nPNI(t)=f(HGB(t),PALB(t),PCR(t),...) nPNI(t) represents the nutritional status assessment at time t, taking into account multiple nutrition-related parameters such as hemoglobin HGB(t), serum albumin PALB(t), and protein degradation rate PCR(t); The prescription adjustment is based on the following formula: If nPNI(t) <nPNIl, Then TPN(t+1)=a(t)·Pr+b(t)·Ft+c(t)·Cb When nPNI(t) is lower than the threshold nPNIl, the nutritional prescription TPN(t+1) for the next day (t+1) needs to be adjusted according to the current nPNI and TPN(t) of the previous day; here Pr, Ft, Cb represent the amount of protein, fat and carbohydrates respectively; coefficients a(t), b(t), c(t) are weight coefficients dynamically fitted according to nPNI(t) and TPN(t), which are used to adjust the proportion of various nutrients.

6. The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology according to claim 1, characterized in that: The full-process closed-loop management module collects all data generated in each link of TPN treatment through the hospital-wide information integration platform, including prescription data, dispensing data, infusion data, and monitoring data, and uses blockchain technology to establish a TPN full-process data chain to ensure the integrity, accuracy, immutability and traceability of the data; the full-process closed-loop management module is based on the five basic principles of 5R, namely Right patient, Right drug, Right dose, Right route, and Right time, to implement TPN full-process closed-loop management; the full-process closed-loop management module intuitively displays the patient's complete TPN treatment history in the form of a timeline, spanning multiple links such as outpatient, hospitalization, and discharge; each node of the timeline can be expanded to present key information, including the time of doctor's order issuance, prescription composition, configuration and adjustment records, drug traceability code, infusion curve, and nutrition monitoring report, which is convenient for full review; each link strictly follows the "SCANNING IN, SCANNING "OUT", referred to as SISO specification, that is: scan the patient's wristband at the beginning of each key link, confirm the patient's identity before executing the doctor's order; at the end of each key link, scan the patient's wristband and the doctor's order again, and submit the execution result to form a closed loop; require two people to sign the key links of doctor's order, drug preparation, infusion configuration, and infusion pipeline verification, and the second signer must be authenticated by fingerprint recognition or iris recognition to eliminate the phenomenon of "signing on behalf of others"; each step of the operation forms a traceable electronic signature; all original documents, including doctor's order, preparation, and infusion orders, are printed on special anti-tampering paper and affixed with a "one-time seal" with a unique traceability code printed on the seal; each document must be fully archived after execution for later reference; the embedded TPN quality continuous improvement submodule can implement T2-1 statistical process control, i.e. SPC, for key indicators such as doctor's order standardization, preparation accuracy, and infusion error rate. According to the intelligent early warning of SPC control chart and continuous optimization, all data are stored in the central database in a unified format and standard, and shared according to the predetermined data access control strategy for use by relevant departments and management departments. The pharmacy department can access prescription and dispensing data to evaluate the rationality of prescriptions and the quality of dispensing; the nursing department can access infusion and monitoring data to evaluate the standardization and safety of infusion management; the medical management department can access the full process data to evaluate the overall diagnosis and treatment quality and efficiency. The full process closed-loop management module is also embedded with data analysis and mining tools to regularly analyze the full process data, generate quality reports, identify potential problems, and put forward continuous improvement suggestions. The continuous improvement suggestions include: analyzing the medical prescription medication pattern, discovering unreasonable prescription patterns and standardizing them; analyzing the clinical efficacy of different prescriptions and optimizing the TPN prescription library; analyzing the relationship between different infusion strategies and the incidence of adverse events, and formulating best infusion practice specifications.

7. The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology according to claim 1, characterized in that: Input age A, gender G, height H, weight W; body composition: muscle mass M, fat mass F, water content H2O; biochemical indicators: albumin ALB, lymphocyte count LY, blood sugar GLU, renal function eGFR, liver function ALTAST; Clinical diagnosis: primary disease D1, complications D2, surgical history SH, infection status INF; The nutritional risk assessment model uses principal component analysis and logistic regression methods: NRS=f(A,G,H,W,M,F,H2O,ALB,LY)=β0+β1PC1+β2PC2+...+β n PC r Among them, PC i is the i-th principal component, β i is the corresponding regression coefficient; the objective function is to minimize NRS while meeting nutrient requirements: Min INRS st Calorie requirement: HC (kcal / d) = BEE × AF × IF × TF Protein requirement: PR (g / kg / d) = IBW × (1.2-2.0) Fat energy supply ratio: 20%~30% Carbohydrate energy supply ratio: 50% to 70% Electrolyte concentration limit: Na + (135-145mmol / L); K + (3.5-5.5mmol / L) Maximum infusion volume: 40mL / kg / d Osmotic pressure limit: 900mOsm / L Drug incompatibility: Incompatible drug pairs {(D1,D2),(D3,D4),...} The sequential quadratic programming algorithm is used to solve the optimal TPN formula: TP={V,AA,LIP,GLU,Na + ,K + ,...} Among them, V is the TPN infusion volume, AA is the amino acid volume, LIP is the fat emulsion volume, and GLU is the glucose volume.

8. The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology according to claim 1, characterized in that: The structuring of medical orders uses a natural language processing algorithm to input the free text of TPN medical orders. The specific steps are as follows: (1) Word segmentation and part-of-speech tagging: Segment the medical order text into words and mark the part of speech of each word; (2) Named entity recognition: identifying key entities such as drug name, dosage number, unit, and route of administration; (3) Relationship extraction: extracting key attribute pairs of drug-dose and drug-administration route; (4) Attribute normalization: Map synonyms, abbreviations, and typos to a standard vocabulary; (5) Five-tuple filling: fill in the standard template of drug, dosage, route of administration, frequency, and course of treatment according to the extracted attribute pairs, and mark the attributes that cannot be filled as null values; Output a structured quintuple consisting of drug, dose, route of administration, frequency, and duration of treatment.

9. The total parenteral nutrition management and control system based on artificial intelligence and Internet of Things technology according to claim 3, characterized in that: The rule-based reasoning includes the following steps: Knowledge base: (1) Drug interaction knowledge base: define a group of if-then rules: If {diazepam, midazolam}∈prescription Then alarm "diazepam and midazolam should not be used at the same time"; (2) Incompatibility database: defined as if-then rules: If {amino acid, fat emulsion}∈prescription&concentration (amino acid)>6%Then alarm "It is not suitable to use fat emulsion when the amino acid concentration is too high"; (3) Dose range library: defined as the value interval: the glucose concentration for children should be between 5% and 12%; Reasoning process: (1) Matching mode: match the structured medical instructions with the rule conditions of the knowledge base; (2) Conflict detection: check whether there is a logical conflict in the conclusion part of the rule; (3) Problem output: output the problem description and revision suggestions of the hit rule.

Images