A transfusion strategy optimization system based on clinical indicators and molecular markers

Abstract

The application relates to the field of medical information technology, and particularly discloses a blood transfusion strategy optimization system based on clinical indexes and molecular markers, which first collects clinical indexes and molecular marker data from circulating tumor cells of patients, and carries out standardized integration to construct a multidimensional feature set reflecting the association between clinical dynamics and molecular characteristics; key association rules are identified through data mining, and a quantitative abnormal expression spectrum and risk stratification label are generated for each patient according to the rules; the blood transfusion urgency state is evaluated in combination with real-time data, physiological responses and risk changes under different blood transfusion schemes are deduced, the optimal individualized blood transfusion strategy parameters are solved, the strategy is deduced and verified through the construction of a patient-specific virtual simulation scene, and the expected performance deviation is calculated by comparing with historical data to output the final execution scheme after optimization and verification; and the application realizes the transformation from static threshold blood transfusion to individualized dynamic prediction and optimized blood transfusion.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, specifically to a transfusion strategy optimization system based on clinical indicators and molecular markers. Background Technology

[0002] Patients with hematologic malignancies such as multiple myeloma and critical illnesses often experience treatment-related or disease-related anemia. Blood transfusion is a crucial supportive treatment for correcting anemia and maintaining tissue oxygenation. Current clinical transfusion practice primarily relies on hemoglobin levels as a core indicator and follows fixed threshold strategies established based on population studies, such as "restrictive transfusion" or "open transfusion." While these strategies provide general guidance, they have significant limitations in individualized application.

[0003] Current technologies lack a way to deeply couple and analyze static, one-off molecular-level heterogeneous information (such as gene expression and mutation profiles of circulating tumor cells) with dynamic, continuous macroscopic physiological indicators (such as hemoglobin levels and organ function scores), and to build a dynamic virtual system based on this coupling relationship that can simulate individualized pathophysiological responses of patients and perform pre-transfusion strategy simulation verification and iterative optimization, thereby changing the current passive response mode of transfusion decision-making that relies on fixed thresholds and lacks individualized prediction and verification capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide a transfusion strategy optimization system based on clinical indicators and molecular markers to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A transfusion strategy optimization system based on clinical indicators and molecular biomarkers includes:

[0007] The data acquisition module is used to acquire patients' clinical indicator data and molecular marker data; the clinical indicator data includes hemoglobin level, organ function score and multiple organ dysfunction status; the molecular marker data includes gene expression profile and gene mutation profile obtained from circulating tumor cells;

[0008] The data integration module is used to receive clinical indicator data and molecular marker data, and to perform structured processing on a patient-by-patient basis to construct a set of associated features containing multi-dimensional characteristics.

[0009] The data analysis module is used to perform data mining and feature extraction on the associated feature set, identify potential association rules between clinical indicators and molecular markers, and generate corresponding abnormal expression profiles and risk stratification labels for each patient based on the potential association rules.

[0010] The strategy generation module is used to output individualized blood transfusion strategies based on abnormal expression profiles and risk stratification labels, combined with the patient's real-time updated clinical indicators and molecular marker dynamic information, through preset reasoning logic. The blood transfusion strategy clearly includes the timing of blood transfusion triggering, hemoglobin threshold and red blood cell transfusion dose.

[0011] The validation and optimization module is used to evaluate the effectiveness and adjust the parameters of individualized blood transfusion strategies. Through simulation and retrospective comparison with historical case data, it analyzes the impact of the strategy on organ function and complications, and finally outputs the optimized blood transfusion execution plan.

[0012] As a further aspect of the present invention: the construction of the associated feature set containing multi-dimensional features specifically includes:

[0013] Clinical indicator data and molecular marker data were preprocessed separately to generate time series matrices of clinical indicators and feature matrices of molecular markers, respectively.

[0014] Based on the preset time window, the trend of indicator changes at multiple consecutive time points in the clinical indicator time series matrix is ​​mapped to the molecular marker feature matrix, generating an association mapping matrix that represents the correspondence between clinical dynamics and molecular static features.

[0015] Based on the association mapping matrix, covariation patterns of clinical trends and molecular characteristics within each time window are extracted, and the covariation patterns are encoded into feature vectors of fixed dimensions. The feature vectors of all time windows are aggregated to form an association feature set.

[0016] As a further aspect of the present invention: the preprocessing of clinical indicator data and molecular marker data specifically includes:

[0017] A sliding window is used to divide the continuous time series of clinical indicator data. The difference calculation is performed on the sequence data in each window to obtain the local rate of change and fluctuation intensity of each clinical indicator. The statistical baseline value in the sliding window is calculated. The rate of change, fluctuation intensity and baseline value are arranged in chronological order to generate a time series matrix of clinical indicators.

[0018] Calculate the local signal-to-noise ratio (SNR) of gene mutation frequency in molecular biomarker data on genomic coordinates, screen out SNR coordinate regions that are higher than a preset threshold, normalize the gene mutation frequency in the SNR coordinate regions, and generate a molecular biomarker feature matrix.

[0019] Based on a unified sampling time point, the time series matrix of clinical indicators and the feature matrix of molecular markers are time-series aligned and dimension-registered to establish a computable mapping relationship between the two on the time axis and feature dimension.

[0020] As a further aspect of the present invention: the rules for identifying potential associations between clinical indicators and molecular biomarkers specifically include:

[0021] On the associated feature set, a parallel search based on multi-level constraints is performed. The multi-level constraints include the physiologically reasonable range of changes in clinical indicators and the biological pathway information of molecular marker expression, in order to discover feature co-occurrence patterns that simultaneously satisfy multiple constraints.

[0022] The co-occurrence patterns of the searched features are subjected to statistical significance tests and causal direction assessments. Patterns with random co-occurrence and reverse causality are eliminated, while patterns with stable leading-lagging time series relationships are retained.

[0023] The quantitative relationships between clinical indicator features and molecular biomarker features in the retained patterns are extracted into a set of association rules with confidence and support.

[0024] As a further aspect of the present invention: the step of generating corresponding abnormal expression profiles and risk stratification labels for each patient based on potential association rules specifically includes:

[0025] Each patient's set of associated features is matched against the set of associated rules one by one, and the deviation of the patient's feature values ​​from each rule is calculated. The deviation is measured by the standardized distance between the patient's feature values ​​and the expected values ​​of the rules.

[0026] The deviation of patients from all association rules is aggregated, and the molecular functions associated with the association rules are hierarchically aggregated to generate an abnormal expression profile that reflects the degree of abnormality of patients in multiple dimensions.

[0027] Based on the severity and combination of abnormalities at different levels in the abnormal expression profile, and referring to a pre-defined risk level mapping table, comprehensive risk stratification labels are assigned to patients.

[0028] As a further aspect of the present invention: the step of outputting an individualized blood transfusion strategy through preset reasoning logic specifically includes:

[0029] Based on risk stratification labels and real-time updated clinical indicators, the real-time rate of decline in hemoglobin levels is calculated and compared with the organ function compensation threshold corresponding to the risk stratification labels to assess the patient's current transfusion urgency.

[0030] By combining the degree of abnormality in molecular function in the abnormal expression profile, we can predict the expected physiological response curve and potential risk changes of patients to blood transfusion within a preset time period.

[0031] With the optimization goal of maximizing tissue oxygenation efficiency and minimizing the risk of transfusion-related complications, under the constraints of transfusion urgency and expected physiological response curves, the optimal hemoglobin threshold, infusion dose and corresponding trigger time point are dynamically solved through iterative calculation to form an individualized transfusion strategy.

[0032] As a further aspect of the present invention: the prediction of the patient's expected physiological response curve and potential risk changes to blood transfusion within a predetermined time period by combining the degree of abnormality in molecular function in the abnormal expression profile specifically includes:

[0033] By analyzing the abnormal expression profile, the deviation of patients from cellular oxygen utilization, iron metabolism homeostasis, and molecular functions related to vascular endothelial response is quantified, and a patient-specific physiological perturbation vector is formed.

[0034] Based on physiological perturbation vectors, the standard coupling relationship between changes in hemoglobin concentration and tissue oxygen uptake efficiency after blood transfusion is personalized and modified to construct the patient's blood transfusion-oxygen supply response function.

[0035] Using the transfusion-oxygen supply response function, we can deduce the dynamic trajectory of the patient's hemoglobin concentration and the corresponding improvement process of tissue oxygenation level after different doses of red blood cells are introduced within a preset time period in the future.

[0036] Simultaneously assess the activation intensity of dynamic change trajectories on known complication pathways associated with patient risk stratification labels, forming a potential risk change curve parallel to hemoglobin changes.

[0037] As a further aspect of the present invention: the final output optimized blood transfusion execution scheme specifically includes:

[0038] Based on the patient's individual set of associated features and abnormal expression profile, a virtual simulation scenario reflecting the patient's pathophysiological characteristics is constructed.

[0039] In a virtual simulation scenario, a personalized blood transfusion strategy is loaded and executed, and the dynamic changes in organ function indicators and the probability of complications are calculated simultaneously during the simulation process.

[0040] The dynamic changes in organ function indicators and the probability of complication are matched and compared with the actual clinical outcome data of cases with similar abnormal expression profiles in the historical case database to calculate the deviation of the strategy from the expected efficacy.

[0041] Based on the expected performance deviation, the transfusion triggering timing, hemoglobin threshold, and red blood cell transfusion dosage parameters in the individualized transfusion strategy are adjusted in a targeted manner to generate an optimized transfusion execution plan.

[0042] As a further aspect of the present invention: the construction of a virtual simulation scene reflecting the patient's pathophysiological characteristics specifically includes:

[0043] Based on the abnormal expression profile, we identify and extract the core molecular functional abnormal events that drive the patient's current physiological state, as well as the key change patterns of these abnormal events in clinical indicators.

[0044] Based on the set of associated features, a network of time dependence and causal strength relationships between core molecular functional abnormality events and key clinical change patterns was constructed.

[0045] Based on a relational network, real-time physiological time-series data of patients are integrated to generate a dynamic virtual simulation scenario that includes multi-level pathophysiological interactions and can simulate the state evolution under blood transfusion intervention.

[0046] The beneficial effects of this invention are:

[0047] (1) This invention systematically integrates time-series clinical indicators and deep molecular biomarker data, and mines their inherent correlation rules to construct individualized abnormal expression profiles and risk stratification labels for each patient. This makes risk assessment no longer limited to general "high risk" or "low risk," but specific to specific molecular dysfunction dimensions. More importantly, by constructing personalized transfusion-oxygen supply response functions and complication risk models, it is possible to predict the physiological effects and potential risks of different transfusion protocols for a specific patient before implementation, thereby solving for the theoretically optimal individualized strategy parameters (including timing, threshold, and dosage). This method overcomes the limitations of traditional "one-size-fits-all" thresholds and helps to improve tissue oxygen supply while minimizing the risk of insufficient or excessive transfusion due to individual patient differences.

[0048] (2) This invention introduces a virtual simulation scenario based on patient-specific pathophysiological characteristics. In this scenario, the system can simulate the execution of recommended individualized transfusion strategies and dynamically extrapolate their cascading effects on organ function and complication risks. By quantitatively comparing the extrapolation results with the actual outcomes of similar historical cases, the "deviation of expected efficacy" of the strategy can be calculated. This indicator provides an objective pre-assessment of the strategy's reliability. Based on this deviation, the system can automatically adjust and optimize the strategy parameters in a directional and iterative manner until it outputs an optimized execution plan that has been verified by simulation and whose expected effect is closest to historical best practices. This closed-loop process not only significantly improves the scientific nature and reliability of decision-making but also enables the system to continuously improve and evolve based on new data. Attached Figure Description

[0049] The invention will now be further described with reference to the accompanying drawings.

[0050] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figure 1 As shown, this invention is a transfusion strategy optimization system based on clinical indicators and molecular markers, comprising:

[0053] The data acquisition module is used to acquire patients' clinical indicator data and molecular marker data; the clinical indicator data includes hemoglobin level, organ function score and multiple organ dysfunction status; the molecular marker data includes gene expression profile and gene mutation profile obtained from circulating tumor cells;

[0054] The data integration module receives clinical indicator data and molecular marker data, and performs standardization and structuring processing on a patient-by-patient basis to construct a set of associated features containing multi-dimensional characteristics.

[0055] The data analysis module is used to perform data mining and feature extraction on the associated feature set, identify potential association rules between clinical indicators and molecular markers, and generate corresponding abnormal expression profiles and risk stratification labels for each patient based on the potential association rules.

[0056] The strategy generation module is used to output individualized blood transfusion strategies based on abnormal expression profiles and risk stratification labels, combined with the patient's real-time updated clinical indicators and molecular marker dynamic information, through preset reasoning logic. The blood transfusion strategy clearly includes the timing of blood transfusion triggering, hemoglobin threshold and red blood cell transfusion dose.

[0057] The validation and optimization module is used to evaluate the effectiveness and adjust the parameters of individualized blood transfusion strategies. Through simulation and retrospective comparison with historical case data, it analyzes the impact of the strategy on organ function and complications, and finally outputs the optimized blood transfusion execution plan.

[0058] In the data acquisition module, clinical indicator data are obtained through routine medical testing and evaluation procedures. Specifically, hemoglobin levels are obtained directly through venous blood sampling and routine blood tests using a fully automated blood analyzer. Organ function scores are calculated using either the clinically accepted Sequential Organ Failure Assessment (SEF) or Acute Physiology and Chronic Health Evaluation (APGE) system. For the SEF, the functional status of six systems—respiratory, coagulation, liver, circulatory, nervous, and renal—is assessed on a scale of 0 to 4, and the scores for each system are summed to obtain the total score. For the APGE, the acute physiological score is first calculated based on the 12 worst physiological parameters within 24 hours of admission to the intensive care unit, and then combined with the patient's age score and chronic health status score; the three are then added together to obtain the total score. The determination of multiple organ dysfunction syndrome (MODS) is based on relevant clinical diagnostic criteria, and is accomplished by assessing and recording the function of the cardiovascular, respiratory, renal, liver, hematologic, and nervous systems item by item.

[0059] Molecular biomarker data were obtained from patient peripheral blood samples using laboratory analytical techniques. First, circulating tumor cells (CTCs) were isolated and enriched from the patient's peripheral blood using density gradient centrifugation. Subsequently, single-cell sequencing analysis was performed on the enriched CTCs: a single-cell RNA sequencing platform was used to amplify and sequence the transcriptome of individual cells, generating a gene expression profile for each cell, presented as a unique molecular identifier count matrix for each gene in each cell; whole-exome sequencing or targeted sequencing panel technology targeting known driver genes was used to sequence the genomic DNA of the CTCs, detecting and generating mutation profiles specific to the genome. All raw sequencing reads underwent quality control processes, including but not limited to filtering low-quality cells, removing low-expression genes, and standardizing gene expression counts, to ensure data quality met the requirements for subsequent analysis.

[0060] In the data integration module, the clinical indicator data and molecular marker data are first preprocessed separately to construct data structures suitable for correlation analysis.

[0061] Clinical indicator data are essentially sequential data collected continuously over time. During processing, a sliding window of fixed length is used to sequentially extract subsequences of data along the time axis. For each window, the numerical sequence of a specific clinical indicator (e.g., hemoglobin level) is differentially calculated. Specifically, the value at the next time point within the window is subtracted from the value at the previous time point, and then divided by the time interval between the two time points to obtain the local rate of change (i.e., instantaneous slope) of the indicator within the window. Simultaneously, the standard deviation of all values ​​for that indicator within the window is calculated as a measure of its variability. Furthermore, the arithmetic mean of all values ​​within the window is calculated as the statistical baseline for that time period. This process is repeated for each clinical indicator, and the "rate of change," "variability," and "baseline value" of all indicators calculated for each time window are arranged chronologically to form a clinical indicator time-series matrix. The rows of this matrix correspond to different time windows, and the columns correspond to different derived characteristics of different indicators.

[0062] For molecular biomarker data, the focus is on filtering high-quality feature regions from massive genomic coordinate information. For gene mutation profile data, the local signal-to-noise ratio (SNR) of mutation frequencies at specific genomic coordinate points (such as exon regions of a gene) is calculated. Specifically, the mutation reading at that coordinate point is divided by the average mutation reading within a pre-defined background region; the resulting ratio is defined as the SNR at that point. A SNR threshold (e.g., a ratio of 3) is set, and all coordinate regions with SNRs higher than this threshold are selected, considered to carry significant biological signals. Subsequently, only the mutation frequency data within these high SNR coordinate regions are retained, and these frequency values ​​are normalized to a uniform range between 0 and 1. The processed data is organized into a molecular biomarker feature matrix, where rows correspond to different patient samples (or cells), and columns correspond to different selected high SNR genomic coordinate features.

[0063] To correlate the two types of heterogeneous data, temporal alignment and dimensional registration are required. Based on the uniform sampling time point during clinical indicator data collection, the corresponding or nearest-neighbor time point identifiers for molecular biomarker sample collection are identified. Using this as a basis, each time window in the clinical indicator time series matrix is ​​associated with the corresponding sample row in the molecular biomarker feature matrix. Simultaneously, it is ensured that the calculations involved in subsequent analyses are consistent across the matrix dimensions.

[0064] Next, an association mapping matrix is ​​constructed. A mapping time window containing multiple consecutive time points (e.g., three) is defined. For each such window, data from the corresponding time row in the clinical indicator time series matrix is ​​extracted, and the trend of each clinical indicator derived feature (such as the rate of change in hemoglobin) within that window is calculated, for example, by calculating the slope of its linear regression. This calculated set of trend values ​​is treated as a whole vector and mapped to the row of the molecular marker feature matrix registered with the central time point of that mapping window. This combination of "trend vector - molecular feature row" constitutes a basic mapping unit. This operation is repeated for all mapping time windows, and the resulting set is the association mapping matrix, which represents the correspondence between dynamic clinical trends and static molecular features at different time points.

[0065] Finally, covariance patterns are extracted based on the association mapping matrix to form the final set of association features. For each mapping unit in the association mapping matrix (i.e., a combination of a trend vector and a row of molecular features), the covariance relationship between the elements in the trend vector and the elements in the molecular feature row is analyzed. For example, the Pearson correlation coefficient between the "hemoglobin change rate trend" in the trend vector and the "specific gene mutation frequency" in the molecular feature row is calculated (first, the arithmetic mean of all values ​​in the "hemoglobin change rate trend" sequence and the arithmetic mean of all values ​​in the "specific gene mutation frequency" sequence are calculated respectively; then, the arithmetic mean of the sequence is subtracted from each value in the "hemoglobin change rate trend" sequence to obtain its deviation value, and the same deviation calculation is performed on the "specific gene mutation frequency" sequence; then, the two deviation values ​​at the same sequential position are multiplied, and all these products are summed to obtain the numerator of the covariance of the two sequences; subsequently, all deviations of the "hemoglobin change rate trend" sequence are calculated respectively). The sum of squares of the values ​​and the sum of squares of all deviations in the "specific gene mutation frequency" sequence are calculated, and the square root of each is taken to obtain the product of the standard deviations of the two sequences as the denominator. Finally, the covariance numerator calculated above is divided by the product of the standard deviations and the denominator, and the quotient is the Pearson correlation coefficient, which has a value between negative one and positive one. The absolute value of this value directly represents the strength of the linear covariance relationship between the two sequences. As a measure of the strength of covariance, all possible trend-feature pairs within a mapping unit are traversed, and the calculated covariance strength values ​​are arranged into a one-dimensional vector in a fixed order (e.g., first sorted by clinical features, then by molecular features). This is the covariance pattern encoding vector for that time window. The covariance pattern encoding vectors generated by all time windows are aggregated, i.e., stacked in chronological order, to form a set of association features containing multi-dimensional features for subsequent analysis.

[0066] In the strategy generation module, based on the abnormal expression profile and risk stratification labels generated in the preceding steps, and combined with the patient's real-time updated clinical indicators and dynamic molecular marker information, a personalized transfusion strategy is output through preset reasoning logic. This strategy is presented in the form of executable parameters, explicitly including the transfusion triggering timing, hemoglobin threshold target, and red blood cell transfusion dose.

[0067] First, the patient's current transfusion urgency is assessed. This assessment relies on risk stratification labels and real-time updated clinical indicators. Specifically, a sequence of hemoglobin levels from several recent consecutive time points (e.g., the last 6 hours) is extracted from real-time clinical indicators, and a linear regression analysis is performed on this sequence to calculate its slope. This slope value, representing the real-time rate of decrease in hemoglobin levels, is expressed in grams per liter per hour (g / L / h). Simultaneously, corresponding organ function compensation thresholds are pre-defined for different risk stratification labels. This threshold is a critical hemoglobin level (in g / L), representing the lowest hemoglobin level that can be tolerated to maintain basic organ function under that risk state. The current hemoglobin level value, and the predicted hemoglobin level value extrapolated to a pre-defined assessment window (e.g., the next 2 hours) based on its rate of decrease, are compared with the organ function compensation threshold corresponding to that risk stratification. If the current level is below the threshold, or the predicted future level will be below the threshold, a transfusion urgency is identified, and its severity level is recorded.

[0068] Secondly, by combining the degree of abnormality in molecular function within the abnormal expression profile, the expected physiological response curve and potential risk changes of the patient to blood transfusion within a predetermined time period are predicted. This process begins with analyzing the abnormal expression profile to quantify the patient's deviation in specific molecular functional dimensions. The abnormal expression profile contains abnormal scores for multiple molecular functional modules (such as cellular oxygen utilization, iron metabolism homeostasis, and vascular endothelial response). By querying a pre-established mapping table based on a biomedical knowledge base, the abnormal score of each functional module is converted into the intensity coefficient and direction (promoting or inhibiting) of its influence on physiological systems (such as tissue oxygen uptake efficiency, iron reserve dynamics, and microvascular reactivity). The influence coefficients of all relevant functional modules are arranged in a predetermined order to form a numerical vector, called the patient-specific physiological perturbation vector. ,in Representing the The quantitative value of each physiological disturbance factor.

[0069] Subsequently, based on this physiological perturbation vector, the "standard coupling relationship between changes in hemoglobin concentration after transfusion and tissue oxygen uptake efficiency" was individually modified. The standard coupling relationship consists of a fundamental function. The description indicates that under standard physiological conditions, when the hemoglobin concentration is... At that time, the corresponding tissue oxygen uptake efficiency. Personalized correction establishes a correction function. Constructing personalized transfusion-oxygenation response functions for patients. The process is defined by the following mathematical formula: ;

[0070] in, This indicates that the patient's hemoglobin concentration is under the current pathophysiological state. The corresponding predicted tissue oxygen uptake efficiency; This represents the standard oxygen uptake efficiency function derived from population physiological data; Represents patient-specific physiological perturbation vector The first in One element; Representative and the The scaling coefficients, which correspond to each perturbation factor and are pre-fitted from experimental data, are used to... The influence is dimensionalized and its intensity is adjusted; Represents a value related to hemoglobin concentration. Related, describing the The characteristic function (e.g., constant, linear function, or saturation function) of the mode of action of each disturbance factor is predefined according to the physiological mechanism; This represents the total number of physiological disturbances considered; the formula transforms the patient's molecular functional abnormality vector into a mathematical description of their individualized oxygen supply response.

[0071] Next, the constructed transfusion-oxygen supply response function is used. The process involves extrapolating the future. A predetermined time period T (e.g., 24 hours) is set. Given a hypothetical red blood cell dose D, the pharmacokinetic model is used to calculate the patient's hemoglobin concentration within the future time period T after this dose input. Over time The dynamic trajectory of change. Pharmacokinetic models include parameters such as volume of distribution and half-life, which can be initially estimated based on individual information such as patient weight and baseline blood volume. Then, Substituting the dynamic trajectory into the personalized function This allows for the calculation of the predicted tissue oxygenation level within the corresponding time period. The process of improvement is a predictable physiological response curve.

[0072] It should be noted that the construction and calculation process of the pharmacokinetic model is as follows: the model inputs include: the planned red blood cell transfusion dose, measured in grams of hemoglobin or units of red blood cells; the patient's individual baseline information, specifically weight, height, and sex, used to estimate their blood volume; and the baseline hemoglobin concentration measured before transfusion. The total blood volume is estimated based on the patient's individual baseline information. For males, the blood volume (in liters) equals 0.03669 multiplied by weight (kg), plus 0.03219 multiplied by height (cm), and then minus 0.6049; for females, the blood volume equals 0.03308 multiplied by weight (kg), plus 0.08569 multiplied by height (cm), and then minus 0.4542. This calculated blood volume value is used as an approximation of the volume of distribution in the pharmacokinetic model. The mean half-life of red blood cells in the body is determined. This value can be set to a standard value (e.g., 30 days) or adjusted according to individualized clinical factors such as the patient's renal function. The elimination rate constant k is calculated based on the half-life, which is the natural logarithm of 2 divided by the half-life. The expected immediate increase in hemoglobin concentration resulting from the infused dose is calculated. This increase is obtained by dividing the dose by the distribution volume. The model outputs a predicted hemoglobin concentration at any future time point t (starting from the moment the transfusion is completed). The specific calculation process is as follows: add the concentration increase multiplied by the negative k of the natural constant e, multiplied by t, to the baseline concentration. Using this method, a continuous dynamic trajectory of hemoglobin concentration can be generated from the moment the transfusion is completed until the end of a predetermined future time period T.

[0073] While projecting physiological responses, the changes in potential complication risks are simultaneously assessed. Specific complication pathways (e.g., circulatory overload, increased thrombotic tendency) are predefined and associated with different risk stratification labels. Each pathway is quantified into a risk scoring function, which is correlated with the current hemoglobin level and its rate of change. On the projected dynamic trajectory of hemoglobin change, the rate of change at each time point is calculated and substituted into the relevant risk scoring functions to calculate the risk score trajectory over time. The risk scores of all relevant pathways are summarized to form a potential risk change curve parallel to the hemoglobin change trajectory.

[0074] It should be noted that the quantification and construction process of the risk scoring function is as follows: The input to the risk scoring function for each complication path is the current hemoglobin level and the rate of change of hemoglobin, and the output is a risk probability value between 0 and 1. The specific form of this function is as follows: First, a linear combination Z is calculated, where Z equals a coefficient *a* related to the hemoglobin level *H* multiplied by *H*, plus a coefficient *b* related to the rate of change *ΔH* multiplied by *ΔH*, plus a constant term *c*. Then, this linear combination Z is substituted into the logistic function, i.e., 1 divided by (1 plus the negative Z power of the natural constant *e*), to obtain the final risk score. The coefficients *a*, *b*, and the constant *c* are obtained through logistic regression analysis of historical case data: The *H* and *ΔH* sequence data of patients who had previously experienced the complication are extracted from the historical database as positive samples, and the data of those who did not experience the complication are used as negative samples. Maximum likelihood estimation is used to train the model, and finally, the specific values ​​of *a*, *b*, and *c* that optimize the model's prediction accuracy are determined, thus completing the quantification of the risk scoring function.

[0075] Finally, with the optimization objective of maximizing tissue oxygenation efficiency and minimizing the risk of transfusion-related complications, the optimal strategy parameters are dynamically solved through iterative calculations under multiple constraints, including the assessed transfusion urgency state, the projected expected physiological response curve, and the potential risk change curve. An objective function is defined, which represents the total increase in tissue oxygenation level (obtained by analyzing the response curve). The weighted sum of the scores (obtained by integrating the risk change curve over the same time period) is subtracted from the weighted sum of the scores for potential risks. The weighting coefficients reflect the different levels of clinical emphasis on benefit versus risk. Constraints include: the transfusion trigger time is no earlier than the current time and no later than the latest time allowed by the urgency assessment; the peak hemoglobin level reached after transfusion does not exceed the safety limit; and the transfusion dose is within the single-transfusion safety range. Numerical optimization methods (such as gradient descent or grid search) are used to iteratively search the strategy parameter space (trigger time, target hemoglobin threshold, transfusion dose) to find the parameter combination that maximizes the above objective function value. This set of parameters is then determined as the individualized optimal transfusion strategy for the patient.

[0076] In the validation and optimization module, the effectiveness of the generated individualized transfusion strategy is evaluated and the parameters are tuned. The core method is to construct a virtual simulation environment that reflects the patient's unique pathophysiological state. In this environment, the strategy execution process is simulated, and the predicted results are compared and analyzed with historical clinical data in the real world, thereby completing the optimization of the strategy.

[0077] The construction of the virtual simulation scenario begins with the analysis of the patient's abnormal expression profile. Based on the scores of each molecular functional module in the abnormal expression profile, the core abnormal events driving the current pathological state are identified. Specifically, an abnormality score threshold is set for each functional module (e.g., a score exceeding 0.8 is considered significantly abnormal), and modules exceeding this threshold are marked as core abnormal events. Simultaneously, from the associated feature set, patterns of change in clinical indicators that are highly covariant in time with these core abnormal events are extracted (e.g., the "iron metabolism abnormality" event is found to be continuously synchronized with the rate of increase in serum ferritin). These extracted patterns are the key clinical change patterns.

[0078] After identifying core abnormal events and key clinical change patterns, it is necessary to quantify their dynamic relationships to construct a relationship network. By analyzing the time-series data of corresponding features in the associated feature set, the time dependence and causal strength of each "core abnormal event-key clinical change pattern" pair are calculated. Time dependence is determined by calculating the maximum cross-correlation between the two events with time delays; causal strength is assessed using Granger causality tests or similar methods to evaluate whether the time-series information of one event helps predict the change of another event, with a p-value of less than 0.05 as the criterion for a significant causal relationship. Connecting all these tested relationships with quantitative strength (e.g., correlation coefficient) and direction (promotion or inhibition) forms a quantitative, weighted "molecular function-clinical phenotype" causal network.

[0079] Based on this causal network, real-time and continuous physiological time-series data of patients are integrated to generate the final dynamic virtual simulation scene. This is achieved by defining each node in the causal network (such as the degree of abnormality in a certain molecular function or a certain clinical indicator value) as a state variable with an initial state (taken from real-time patient data). The quantified relationship of each edge in the network is transformed into a mathematical function, which defines the rules governing the influence of changes in the state of the source node on the state of the target node. In this way, the entire network constitutes a multivariate, interacting dynamic system model. By running this system model, the chain evolution of all state variables (including organ function indicators) in the system over time after an external intervention (such as blood transfusion) can be simulated.

[0080] In this virtual simulation scenario, a personalized transfusion strategy to be evaluated is loaded and executed. During the simulation, parameters such as the transfusion trigger time and infusion dose defined in the strategy are used as external input events to trigger immediate changes in corresponding state variables such as blood volume and hemoglobin concentration in the scenario. Subsequently, the system continuously calculates according to preset causal rules and outputs the dynamic changes of all organ function indicators (such as the curve of stroke volume changing with time) and the probability curve of complication occurrence calculated by a preset complication risk function throughout the simulation period.

[0081] After the simulation is completed, the calculated organ function improvement (e.g., area under the curve) and peak complication probability are compared with the actual clinical outcomes (e.g., actual survival rate, actual complication rate) of cases with highly similar abnormal expression profiles in the historical case database. The matching process uses the nearest neighbor algorithm, which minimizes the Euclidean distance of the scores of each dimension of the abnormal expression profile to find the most similar historical case group. The specific method for calculating the deviation of the strategy's expected efficacy is as follows: the relative error between the predicted organ function improvement value and the actual average improvement value of the similar case group, and the relative error between the predicted complication probability and the actual average incidence rate are calculated separately. These two relative errors are then weighted and summed (with weights preset to 0.7 and 0.3) to obtain a comprehensive expected efficacy deviation value.

[0082] Finally, based on the calculated deviation from the expected efficacy, the parameters of the individualized transfusion strategy are adjusted in a targeted manner. If the deviation is positive, indicating that the strategy may be too aggressive, the hemoglobin threshold is increased by a fixed step size (e.g., 0.5 units) or proportionally to the magnitude of the deviation, or the transfusion dose is decreased. If the deviation is negative, the adjustment is reversed. This adjustment process can be iterative, i.e., a new strategy is generated using the adjusted parameters, and simulations and comparisons are performed again until the absolute value of the deviation from the expected efficacy is lower than a preset convergence threshold (e.g., less than 0.05). Ultimately, the set of strategy parameters that minimizes the deviation from the expected efficacy is determined as the optimized and validated transfusion execution plan and output.

[0083] The working principle of this invention is as follows: First, it collects multi-dimensional clinical indicator data from patients, such as hemoglobin levels and organ function scores, as well as molecular marker data, including gene expression profiles and mutation profiles, obtained by sequencing circulating tumor cells. Second, it performs separate preprocessing and standardized integration of the two types of data. By constructing an association mapping matrix and extracting covariation patterns between clinical dynamic trends and molecular characteristics, a multi-dimensional association feature set is formed. Third, based on this feature set, it mines potential association rules between clinical indicators and molecular markers, and generates quantified abnormal expression profiles and risk stratification labels for each patient. Then, combining real-time updated clinical data and molecular information, it assesses the urgency of transfusion and uses individualized transfusion-oxygen supply response functions and pharmacokinetic models to deduce physiological response and complication risk curves under different transfusion strategies. Finally, it uses optimization algorithms to solve for the optimal individualized transfusion strategy, including triggering timing, hemoglobin threshold, and infusion dose. Finally, a virtual simulation scenario reflecting the patient's specific pathophysiological characteristics is constructed. In this scenario, a strategy is loaded and deduced. The deduction results are compared with the actual outcomes of similar historical cases to calculate the expected efficacy deviation. Based on this, the strategy parameters are iteratively adjusted until an optimized and validated blood transfusion execution plan is output.

[0084] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A transfusion strategy optimization system based on clinical indicators and molecular biomarkers, characterized in that, include: The data acquisition module is used to acquire patients' clinical indicator data and molecular marker data; the clinical indicator data includes hemoglobin level, organ function score and multiple organ dysfunction status; the molecular marker data includes gene expression profile and gene mutation profile obtained from circulating tumor cells; The data integration module receives clinical indicator data and molecular marker data, and performs structured processing on a patient-by-patient basis to construct a set of associated features containing multi-dimensional characteristics, specifically including: Clinical indicator data and molecular marker data were preprocessed separately to generate time series matrices of clinical indicators and feature matrices of molecular markers, respectively. Based on the preset time window, the trend of indicator changes at multiple consecutive time points in the clinical indicator time series matrix is ​​mapped to the molecular marker feature matrix, generating an association mapping matrix that represents the correspondence between clinical dynamics and molecular static features. Based on the association mapping matrix, the covariation patterns of clinical trends and molecular features within each time window are extracted, the covariation patterns are encoded into feature vectors of fixed dimensions, and the feature vectors of all time windows are aggregated to form an association feature set. The data analysis module is used to perform data mining and feature extraction on the associated feature set, identify potential association rules between clinical indicators and molecular markers, and generate corresponding abnormal expression profiles and risk stratification labels for each patient based on the potential association rules. The strategy generation module is used to output individualized blood transfusion strategies based on abnormal expression profiles and risk stratification labels, combined with the patient's real-time updated clinical indicators and molecular marker dynamic information, through preset reasoning logic. The blood transfusion strategy clearly includes the timing of blood transfusion triggering, hemoglobin threshold and red blood cell transfusion dose. The validation and optimization module is used to evaluate the effectiveness and adjust parameters of individualized transfusion strategies. Through simulation and retrospective comparison with historical case data, it analyzes the impact of the strategy on organ function and complications, and finally outputs an optimized transfusion execution plan, which includes: Based on the patient's individual set of associated features and abnormal expression profile, a virtual simulation scenario reflecting the patient's pathophysiological characteristics is constructed. In a virtual simulation scenario, a personalized blood transfusion strategy is loaded and executed, and the dynamic changes in organ function indicators and the probability of complications are calculated simultaneously during the simulation process. The dynamic changes in organ function indicators and the probability of complication are matched and compared with the actual clinical outcome data of cases with similar abnormal expression profiles in the historical case database to calculate the deviation of the strategy's expected efficacy. Based on the expected performance deviation, the transfusion triggering timing, hemoglobin threshold, and red blood cell transfusion dosage parameters in the individualized transfusion strategy are adjusted in a targeted manner to generate an optimized transfusion execution plan; The construction of the virtual simulation scenario reflecting the patient's pathophysiological characteristics specifically includes: Based on the abnormal expression profile, we identify and extract the core molecular functional abnormal events that drive the patient's current physiological state, as well as the key change patterns of these abnormal events in clinical indicators. Based on the set of associated features, a network of time dependence and causal strength relationships between core molecular functional abnormality events and key clinical change patterns was constructed. Based on a relational network, real-time physiological time-series data of patients are integrated to generate a dynamic virtual simulation scenario that includes multi-level pathophysiological interactions and can simulate the state evolution under blood transfusion intervention.

2. The transfusion strategy optimization system based on clinical indicators and molecular markers according to claim 1, characterized in that, The separate preprocessing of clinical indicator data and molecular marker data specifically includes: A sliding window is used to divide the continuous time series of clinical indicator data. The difference calculation is performed on the sequence data in each window to obtain the local rate of change and fluctuation intensity of each clinical indicator. The statistical baseline value in the sliding window is calculated. The rate of change, fluctuation intensity and baseline value are arranged in chronological order to generate a time series matrix of clinical indicators. Calculate the local signal-to-noise ratio (SNR) of gene mutation frequency in molecular biomarker data on genomic coordinates, screen out SNR coordinate regions that are higher than a preset threshold, normalize the gene mutation frequency in the SNR coordinate regions, and generate a molecular biomarker feature matrix. Based on a unified sampling time point, the time series matrix of clinical indicators and the feature matrix of molecular markers are time-series aligned and dimension-registered to establish a computable mapping relationship between the two on the time axis and feature dimension.

3. The transfusion strategy optimization system based on clinical indicators and molecular markers according to claim 1, characterized in that, The rules for identifying potential associations between clinical indicators and molecular biomarkers specifically include: On the associated feature set, a parallel search based on multi-level constraints is performed. The multi-level constraints include the physiologically reasonable range of changes in clinical indicators and the biological pathway information of molecular marker expression, in order to discover feature co-occurrence patterns that simultaneously satisfy multiple constraints. The co-occurrence patterns of the searched features are subjected to statistical significance tests and causal direction assessments. Patterns with random co-occurrence and reverse causality are eliminated, while patterns with stable leading-lagging time series relationships are retained. The quantitative relationships between clinical indicator features and molecular biomarker features in the retained patterns are extracted into a set of association rules with confidence and support.

4. The transfusion strategy optimization system based on clinical indicators and molecular markers according to claim 1, characterized in that, The process of generating corresponding abnormal expression profiles and risk stratification labels for each patient based on potential association rules specifically includes: Each patient's set of associated features is matched against the set of associated rules one by one, and the deviation of the patient's feature values ​​from each rule is calculated. The deviation is measured by the standardized distance between the patient's feature values ​​and the expected values ​​of the rules. The deviation of patients from all association rules is aggregated, and the molecular functions associated with the association rules are hierarchically aggregated to generate an abnormal expression profile that reflects the degree of abnormality of patients in multiple dimensions. Based on the severity and combination of abnormalities at different levels in the abnormal expression profile, and referring to a pre-defined risk level mapping table, comprehensive risk stratification labels are assigned to patients.

5. The transfusion strategy optimization system based on clinical indicators and molecular markers according to claim 1, characterized in that, The individualized blood transfusion strategy output through preset reasoning logic specifically includes: Based on risk stratification labels and real-time updated clinical indicators, the real-time rate of decline in hemoglobin levels is calculated and compared with the organ function compensation threshold corresponding to the risk stratification labels to assess the patient's current transfusion urgency. By combining the degree of abnormality in molecular function in the abnormal expression profile, we can predict the expected physiological response curve and potential risk changes of patients to blood transfusion within a preset time period. With the optimization goal of maximizing tissue oxygenation efficiency and minimizing the risk of transfusion-related complications, under the constraints of transfusion urgency and expected physiological response curves, the optimal hemoglobin threshold, infusion dose and corresponding trigger time point are dynamically solved through iterative calculation to form an individualized transfusion strategy.

6. The transfusion strategy optimization system based on clinical indicators and molecular markers according to claim 5, characterized in that, The method of predicting the expected physiological response curve and potential risk changes of patients to blood transfusion within a preset time period by combining the degree of abnormality in molecular function in the abnormal expression profile includes: By analyzing the abnormal expression profile, the deviation of patients from cellular oxygen utilization, iron metabolism homeostasis, and molecular functions related to vascular endothelial response is quantified, and a patient-specific physiological perturbation vector is formed. Based on physiological perturbation vectors, the standard coupling relationship between changes in hemoglobin concentration and tissue oxygen uptake efficiency after blood transfusion is personalized and modified to construct the patient's blood transfusion-oxygen supply response function. Using the transfusion-oxygen supply response function, we can deduce the dynamic trajectory of the patient's hemoglobin concentration and the corresponding improvement process of tissue oxygenation level after different doses of red blood cells are introduced within a preset time period in the future. Simultaneously assess the activation intensity of dynamic change trajectories on known complication pathways associated with patient risk stratification labels, forming a potential risk change curve parallel to hemoglobin changes.

Images