Infusion pump medicine dosage control method and system fused with clinical decision support

Abstract

The invention relates to the field of control, in particular to an infusion pump drug dosage control method and system fusing clinical decision support, and the method comprises the steps: monitoring at least two physiological parameters in real time, including a main treatment index and a side effect precursor index; for each parameter, calculating the parameter credibility based on the parameter signal fluctuation characteristic and the deviation degree of the parameter signal fluctuation characteristic and the predicted value of the PK-PD model, and during state updating, weighting the monitoring data by using the parameter credibility to correct the state of the patient; in the constraint conditions, a multi-dimensional safety state space formed by the main treatment indexes and the side effect precursor indexes is defined, and boundaries are adjusted according to the credibility of the side effect precursor indexes; in the cost function, the length of a prediction time window and the asymmetric penalty weight of a prediction track deviating from a target are adjusted according to the credibility of main treatment indexes, a model prediction control problem is solved to obtain an optimal administration dosage sequence, and a first control instruction is output to an infusion pump to be executed.

Description

Infusion pump drug dosage control method and system integrating clinical decision support Technical Field

[0001] This application belongs to the field of control, and in particular relates to a method and system for controlling drug dosage in an infusion pump that integrates clinical decision support. Background Technology

[0002] Infusion pumps, as key devices for precise and continuous drug delivery in modern medicine, are widely used in various clinical scenarios such as anesthesia, intensive care, chemotherapy, and diabetes management. To achieve personalized treatment, closed-loop control infusion pump systems have emerged. These systems monitor one or more physiological parameters of the patient in real time and automatically adjust the drug delivery rate according to a preset control algorithm, thereby maintaining the patient's physiological state within a target range. Among these, control methods based on the PK-PD model are currently a hot research topic. This model can represent the process of drug concentration changes in the body and its relationship with physiological effects. Control strategies based on this model, such as proportional-integral-derivative control and model predictive control, can predict the patient's future state and optimize drug delivery decisions, improving the safety and effectiveness of treatment compared to traditional open-loop drug delivery methods. However, physiological parameter monitoring signals are easily affected by factors such as patient movement, equipment artifacts, and electromagnetic interference in the clinical environment, leading to noise, drift, or even temporary data loss, and often resulting in unreliable signal quality. Existing control systems struggle to reliably distinguish between real physiological fluctuations and signal interference, potentially causing the controller to respond incorrectly based on distorted data, leading to the risk of overdosing or underdosing. Furthermore, when the monitoring signals used for early warning of side effects are not highly reliable, misjudgments may lead to excessive restrictions on drug administration, affecting treatment efficacy. Adjustments are difficult to make based on the real-time quality of monitoring data and changes in patient condition, thus limiting further optimization of control performance. Summary of the Invention

[0003] This invention proposes an infusion pump drug dosage control method integrating clinical decision support to address the problem that existing technologies struggle to reliably distinguish between real physiological fluctuations and signal interference, and cannot adjust dosages based on real-time monitoring data quality and changes in patient condition. The method includes the following steps: acquiring real-time monitoring data for at least two physiological parameters, including a primary therapeutic indicator and at least one precursor to side effects; calculating the reliability of each physiological parameter based on signal fluctuation characteristics and deviation from the predicted value of a PK-PD model; and constructing and solving a model predictive control problem based on the calculated reliability of each physiological parameter to determine the optimal dosage. The model predictive control problem is achieved through… The parameter confidence level is utilized in the following manner: In state updates, the real-time monitoring data is weighted using the parameter confidence level to correct the patient state used for model prediction; in constraints, a multidimensional safe state space is constructed, defined by the primary treatment indicator and the pre-side effect indicator, and the boundary of the safe state space is adjusted according to the parameter confidence level corresponding to the pre-side effect indicator; in the cost function, key parameters affecting the control strategy are adjusted according to the parameter confidence level corresponding to the primary treatment indicator, including the prediction time window length and the asymmetric penalty weight for deviation of the predicted trajectory from the target; the first control command of the optimal dosing sequence obtained by solving is output to the infusion pump for execution.

[0004] Optionally, the step of calculating parameter reliability based on signal fluctuation characteristics and deviation from the predicted value of the PK-PD model includes: calculating the signal fluctuation index of the physiological parameter and the deviation index from the predicted value of the model; and performing weighted fusion based on the signal fluctuation index and the deviation index to calculate the parameter reliability, wherein the smaller the signal fluctuation and deviation, the higher the parameter reliability.

[0005] Optionally, in the state update, the real-time monitoring data is weighted using the parameter confidence level to correct the patient state used for model prediction, including: using a Kalman filter for state update, wherein the real-time monitoring data is used as observations; and the diagonal elements of the measurement noise covariance matrix R in the Kalman filter are... , set as the confidence level with respect to the corresponding physiological parameter i Inversely proportional.

[0006] Optionally, adjusting the boundary of the safe state space based on the reliability of the parameters corresponding to the side effect precursor indicators includes: when the reliability of the parameters of a certain side effect precursor indicator decreases, shrinking the safe boundary corresponding to the reliability of the parameters towards a preset safe baseline value, wherein the amount of shrinkage is proportional to the degree of decrease in the reliability of the parameters.

[0007] Optionally, adjusting the key parameters affecting the control strategy based on the reliability of the parameters corresponding to the main treatment indicator, wherein the key parameters include the prediction time window length, includes: setting the prediction time window length to a corresponding discrete value based on the preset interval to which the reliability of the parameters of the main treatment indicator belongs, wherein the higher the reliability of the parameters, the longer the prediction time window length is set.

[0008] Optionally, adjusting the key parameters affecting the control strategy based on the reliability of the parameters corresponding to the main treatment indicator, wherein the key parameters include the prediction time window length and the asymmetric penalty weight for deviation of the predicted trajectory from the target, includes: in the cost function, when the predicted value of the main treatment indicator is lower than the lower limit of the target range or higher than the upper limit of the target range, the penalty weight is positively correlated with the reliability of the parameters of the main treatment indicator.

[0009] Optionally, the multidimensional safety state space defined by the primary treatment indicator and the side effect precursor indicator includes: the primary treatment indicator is mean arterial pressure; the side effect precursor indicators include heart rate and stroke volume variability; the safety state space is defined as a set that satisfies all of the following conditions: the mean arterial pressure is within a preset target range; the values ​​of the heart rate and stroke volume variability are within their respective adjusted safety boundaries.

[0010] Furthermore, this invention also relates to an infusion pump drug dosage control system integrating clinical decision support, comprising the following modules: a calculation module for acquiring real-time monitoring data of at least two physiological parameters, said physiological parameters including a primary therapeutic indicator and at least one precursor to side effects, and for each said physiological parameter, calculating parameter reliability based on signal fluctuation characteristics and deviation from the predicted value of a PK-PD model; and a construction module for constructing and solving a model predictive control problem to determine the optimal drug dosage based on the calculated parameter reliability of each of the said physiological parameters, wherein the model predictive control problem utilizes the parameter reliability in the following manner: during state updates, utilizing... The real-time monitoring data is weighted using the reliability of the parameters to correct the patient state used for model prediction; in the constraints, a multidimensional safe state space is constructed, defined by the primary treatment indicator and the pre-side effect indicator, and the boundary of the safe state space is adjusted according to the reliability of the parameters corresponding to the pre-side effect indicator; in the cost function, the key parameters affecting the control strategy are adjusted according to the reliability of the parameters corresponding to the primary treatment indicator, the key parameters including the prediction time window length and the asymmetric penalty weight for deviation of the predicted trajectory from the target; the output module is used to output the first control command of the optimal dosing sequence obtained by solving to the infusion pump for execution.

[0011] Preferably, the step of calculating parameter reliability based on signal fluctuation characteristics and deviation from the predicted value of the PK-PD model includes: calculating the signal fluctuation index of the physiological parameter and the deviation index from the predicted value of the model; and performing weighted fusion based on the signal fluctuation index and the deviation index to calculate the parameter reliability, wherein the smaller the signal fluctuation and deviation, the higher the parameter reliability.

[0012] Preferably, in the state update, the real-time monitoring data is weighted using the parameter confidence level to correct the patient state used for model prediction, including: using a Kalman filter for state update, wherein the real-time monitoring data is used as observations; and the diagonal elements of the measurement noise covariance matrix R in the Kalman filter are... , set as the confidence level with respect to the corresponding physiological parameter i Inversely proportional.

[0013] Preferably, adjusting the boundary of the safe state space based on the reliability of the parameters corresponding to the side effect precursor indicators includes: when the reliability of the parameters of a certain side effect precursor indicator decreases, shrinking the safe boundary corresponding to the reliability of the parameters towards a preset safe baseline value, wherein the amount of shrinkage is proportional to the degree of decrease in the reliability of the parameters.

[0014] Preferably, the step of adjusting the key parameters affecting the control strategy based on the reliability of the parameters corresponding to the main treatment indicators includes the prediction time window length, which includes setting the prediction time window length to a corresponding discrete value based on the preset interval to which the reliability of the parameters of the main treatment indicators belongs, wherein the higher the reliability of the parameters, the longer the prediction time window length is set.

[0015] Preferably, the step of adjusting the key parameters affecting the control strategy based on the reliability of the parameters corresponding to the main treatment indicator, wherein the key parameters include the prediction time window length and the asymmetric penalty weight for deviation of the predicted trajectory from the target, includes: in the cost function, when the predicted value of the main treatment indicator is lower than the lower limit of the target range or higher than the upper limit of the target range, the penalty weight is positively correlated with the reliability of the parameters of the main treatment indicator.

[0016] Preferably, the multidimensional safety state space defined by the primary treatment indicator and the side effect precursor indicator includes: the primary treatment indicator is mean arterial pressure; the side effect precursor indicators include heart rate and stroke volume variability; the safety state space is defined as a set that satisfies all of the following conditions: the mean arterial pressure is within a preset target range; the values ​​of the heart rate and stroke volume variability are within their respective adjusted safety boundaries.

[0017] This invention improves the safety of closed-loop drug dosage control by establishing a reliability assessment mechanism based on real-time monitored physiological parameter data. It reliably distinguishes between genuine physiological fluctuations and signal interference, prioritizing reliable data when correcting patient status, thus obtaining accurate patient status estimates. Regarding safety assurance, the boundaries of the multi-dimensional safety space are adjusted accordingly based on the reliability of side effect warning signals. This avoids excessive restrictions on drug administration due to unreliable warning signals, which could negatively impact treatment efficacy, while also enabling protective measures to be taken when warning signals are indeed reliable. Simultaneously, key parameters in the control strategy can be adjusted based on the data quality of treatment indicators, making the controller's behavior more aligned with the current data situation. In a clinical monitoring environment, this achieves a balance between treatment efficacy and risk control. Attached Figure Description

[0018] Figure 1 is a flowchart of the first embodiment; Figure 2 is a schematic diagram of parameter confidence; Figure 3 is a schematic diagram of the relationship between confidence and the R value of the Kalman filter; Figure 4 is a schematic diagram of security boundary adjustment; Figure 5 is a schematic diagram of prediction time window length adjustment; Figure 6 is a schematic diagram of penalty weight adjustment; Figure 7 is a schematic diagram of multidimensional security state space. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the first embodiment, the present invention proposes an infusion pump drug dosage control method integrating clinical decision support, as shown in Figure 1, comprising the following steps: S1, acquiring real-time monitoring data of at least two physiological parameters, including a primary therapeutic indicator and at least one precursor to side effects, and calculating the reliability of each physiological parameter based on signal fluctuation characteristics and deviation from the predicted value of the PK-PD model; specifically, taking propofol infusion during general anesthesia as an example, the bispectral index (BIS) of the electroencephalogram (EEG) is acquired as the primary therapeutic indicator using a bispectral index monitor, and the mean arterial pressure (MAP) is acquired as a precursor to side effects using an electrocardiogram (ECG) monitor. For the BIS, the signal variance of the index over the past minute is calculated to represent the fluctuation characteristics; the larger the variance, the more unstable the signal. Simultaneously, the currently monitored real-time BIS value is compared with the expected BIS value predicted based on the PK-PD model, and the absolute value of the difference between the two is calculated as the deviation. The PK-PD model is a model of the relationship between drug administration regimens, changes in drug concentration in the patient's body, and the resulting pharmacological effects. It can employ linear, log-linear, maximum-effect, or β-function models, and includes a pharmacokinetic (PK) component to represent the absorption, distribution, metabolism, and excretion of the drug in the patient's body, establishing the relationship between the administered dose and changes in drug concentration over time. The pharmacodynamic (PD) component represents the causal relationship between drug concentration in the body and clinical efficacy or adverse reaction indicators. The input to the PK-PD model is control variables such as the administered dose or infusion rate; some PK-PD models can also input individualized patient parameters. The model output is the predicted values ​​for key therapeutic indicators, side effect-related indicators, and physiological parameters.

[0021] The signal variance and deviation are weighted and fused, and then converted into a confidence value between 0 and 1 using a preset mapping function, such as a negative exponential function. The more stable the signal and the closer it is to the model prediction, the closer the value is to 1, and vice versa. The same method is used to calculate the confidence value for the MAP parameter.

[0022] In an optional embodiment, the step of calculating parameter reliability based on signal fluctuation characteristics and deviation from the predicted values ​​of the PK-PD model includes: calculating the signal fluctuation index of the physiological parameter and the deviation index from the predicted values ​​of the model; and performing a weighted fusion of the signal fluctuation index and the deviation index to calculate the parameter reliability, wherein the smaller the signal fluctuation and deviation, the higher the parameter reliability.

[0023] Data quality is assessed by representing the intrinsic stability of physiological signals and their consistency with model predictions. Signal variability indices are calculated, for example, for a given parameter, based on N data points collected over the past minute. The volatility index V can be calculated as the standard deviation of the data set. If the mean arterial pressure reading sequence at a certain moment is 65, 66, 65, 67, 66 mmHg, the volatility index is very low; if the sequence is 65, 80, 62, 78, 63 mmHg, the volatility index is very high, indicating that the signal may be disturbed and unstable.

[0024] The deviation index is calculated to represent the difference between the real-time monitored value and the predicted value from the patient's pharmacokinetic and pharmacodynamic models. If the model predicts the current mean arterial pressure to be 70 mmHg, while the real-time monitored value is 69 mmHg, the deviation is small. If the monitored value is 85 mmHg, the deviation is large, which may indicate sensor error or an unexpected change in the patient's condition. The deviation index D can be defined as the normalized absolute error. Normalized volatility indicator and deviation index The confidence level C is obtained by weighted fusion, for example, using the formula... ,in and The preset weights can be adjusted according to clinical needs. When both fluctuation and deviation are small, the confidence level C is close to 1, indicating that the data is highly reliable; conversely, it indicates that the data is not reliable, as shown in Figure 2.

[0025] S2, Based on the calculated parameter confidence of each of the physiological parameters, a model predictive control problem is constructed and solved to determine the optimal drug dosage. The model predictive control problem utilizes the parameter confidence in the following ways: In state updates, the real-time monitoring data is weighted using the parameter confidence to correct the patient state used for model prediction; in constraints, a multidimensional safe state space defined by the primary treatment indicator and the side effect precursor indicator is constructed, and the boundary of the safe state space is adjusted according to the parameter confidence corresponding to the side effect precursor indicator; in the cost function, key parameters affecting the control strategy are adjusted according to the parameter confidence corresponding to the primary treatment indicator, the key parameters including the prediction time window length and the asymmetric penalty weight for deviation of the predicted trajectory from the target. The model predictive control problem (MPC) includes a system state prediction model, a set of constraints, and a cost function. During the control process, the patient's state estimated at the current moment is used as the initial condition. Combined with the PK-PD model, the system state and clinical indicators under different candidate drug doses are predicted in a rolling manner within a future prediction time window. Based on this, an optimization problem is constructed that includes treatment target deviation penalty, dose change smoothness and safety constraints. Under the premise of satisfying the multidimensional safety state space constraints, a set of optimal drug dose sequences that minimize the cost function is obtained.

[0026] Specifically, a Kalman filter is used for patient state estimation. In the Kalman filter update step, the measurement noise covariance matrix R is a key parameter, reflecting the uncertainty of the measurement data. An inverse relationship is established between parameter confidence and the R value. When the confidence of the BIS signal is high, for example, close to 1, it indicates that the measurement data is reliable, and the corresponding element value of the BIS in the R matrix is ​​adjusted downwards, making the filter rely more on the current measurement value to correct the patient state. Conversely, when the confidence of the BIS signal is low, for example, close to 0, the R value is increased, making the filter rely more on the predictions of the internal model, reducing the impact of unreliable data on state estimation.

[0027] In anesthesia cases, the safety space is jointly defined by BIS and MAP. BIS needs to be maintained within the target range of 40-60 to ensure appropriate depth of anesthesia, while MAP needs to be no less than 65 mmHg to prevent the risk of hypotension. When the reliability of the MAP monitoring signal decreases due to sensor interference or changes in patient position, the system determines that the reliability of the current blood pressure reading is insufficient. If the rigid constraint of 65 mmHg is strictly enforced, it may lead to excessive reduction of anesthetic drug infusion due to misjudgment, resulting in the risk of insufficient depth of anesthesia. Therefore, the system will automatically and dynamically shrink the safety boundary towards a more conservative baseline value. For example, the lower limit of MAP safety may be temporarily raised from 65 mmHg to 70 mmHg to reduce the impact of unreliable data on control decisions by tightening the constraint range. Conversely, if the reliability of the MAP signal remains high and the blood pressure continues to be stable, the original boundary setting of 65 mmHg will be maintained.

[0028] Regarding the prediction time window length, when the reliability of the primary therapeutic indicator (BIS) is high, it indicates that the current system state is clear and the model prediction is relatively accurate. In this case, a longer prediction time window is set, such as 10 minutes, to allow for longer-term planning and more stable control. If the BIS signal reliability is low, the prediction time window is shortened, for example, to 2 minutes, so that the controller focuses on near-term safety and stability, avoiding long-term predictions based on unreliable data. Regarding asymmetric penalty weights, the risk of excessive anesthesia (BIS below 40) is higher than that of insufficient anesthesia (BIS above 60). When the BIS signal reliability is high, a penalty weight much greater than that above 60 can be applied to predicted trajectories below 40. However, when the BIS signal reliability is low, the authenticity of the deviation direction cannot be determined. In this case, the asymmetry of the penalty weights is reduced to make them more symmetrical, thereby taking more conservative control actions to prevent overdose due to erroneous signals.

[0029] In an optional embodiment, the step of weighting the real-time monitoring data using the parameter confidence level to correct the patient state used for model prediction during state update includes: performing state update using a Kalman filter, wherein the real-time monitoring data is used as observations; and adjusting the diagonal elements of the measurement noise covariance matrix R in the Kalman filter. , set as the confidence level with respect to the corresponding physiological parameter i Inversely proportional, that is ,in This is a preset scaling factor.

[0030] A Kalman filter is employed as a state observer to fuse model predictions and actual measurements, resulting in a more accurate estimate of the patient's true physiological state. The core of this filter lies in balancing the weights between model predictions and sensor observations through Kalman gain. The calculation of the Kalman gain depends on two key parameters: process noise covariance Q and measurement noise covariance R. The former represents the uncertainty of the model prediction, while the latter represents the uncertainty of the sensor measurement.

[0031] In practice, the measurement noise covariance matrix R is a diagonal matrix, and the i-th element on the diagonal... This represents the measurement noise variance of the i-th physiological parameter. This value correlates with the confidence level of that parameter. They are inversely proportional. For example, for the mean arterial pressure parameter, the corresponding measurement noise variance is... Assuming a preset proportional coefficient The confidence level is 0.5. When the mean arterial pressure signal is stable and consistent with the model prediction, the confidence level is [value missing]. The calculation is 0.9, at this point A value of approximately 0.56 is relatively small, allowing the Kalman filter to place more trust in the current measurement when updating the state. Conversely, if the sensor is disturbed, the reliability may be compromised. If it drops to 0.2, then It will increase to 2.5, making the filter rely more on the state predicted by the internal model, while reducing the impact of abnormal measurement data on state estimation, as shown in Figure 3.

[0032] In an optional embodiment, adjusting the boundary of the safe state space based on the reliability of the parameters corresponding to the side effect precursor indicators includes: when the reliability of the parameters of a certain side effect precursor indicator decreases, shrinking the safe boundary corresponding to the reliability of the parameters towards a preset safe baseline value, wherein the amount of shrinkage is proportional to the degree of decrease in the reliability of the parameters.

[0033] To ensure medication safety, the model predictive control algorithm must ensure that all predicted precursors to side effects remain within a preset safety boundary during optimization. This safety boundary is no longer fixed but adjusted based on the data reliability of the corresponding parameters. For each precursor to a side effect, such as heart rate, a standard safety range is defined, for example, 50 to 110 beats per minute, and a more conservative baseline safety range, for example, 60 to 100 beats per minute. The standard range is used when data quality is good, while the baseline range serves as a safeguard when data quality is extremely poor.

[0034] The adjustment of the boundary is directly related to the parameter confidence level C. Let the conventional upper limit be... The baseline is Then the upper limit It can be derived from the formula The calculations show that the lower boundary follows the same logic. For example, with heart rate, when the signal quality is high, the reliability is... When the value is 1.0, the safe range is the standard 50 to 110 beats per minute. If noise in the heart rate signal is caused by patient movement, the reliability... When the value drops to 0.4, the upper safety limit contracts to 104 beats per minute, and the lower limit contracts to 56 beats per minute. The safe range is thus tightened to 56 to 104 beats per minute. This contraction mechanism allows the system to adopt a more conservative control strategy when it cannot accurately monitor a certain indicator, avoiding placing the patient in a potentially dangerous state due to inaccurate data, as shown in Figure 4.

[0035] In an optional embodiment, adjusting the key parameters affecting the control strategy based on the reliability of the parameters corresponding to the main treatment indicator, wherein the key parameters include the prediction time window length, includes: setting the prediction time window length to a corresponding discrete value based on the preset interval to which the reliability of the parameters of the main treatment indicator belongs, wherein the higher the reliability of the parameters, the longer the prediction time window length is set.

[0036] The performance of model predictive control largely depends on the length of the prediction time window. A longer time window allows the controller to plan further ahead, potentially leading to better control performance, but this requires highly accurate models and data. When data quality deteriorates, long-term prediction errors accumulate and amplify, potentially resulting in erroneous control decisions. This can be addressed by correlating the prediction time window length with the confidence level of key therapeutic indicators, such as mean arterial pressure. The adjustments are made by combining these methods, as shown in Figure 5. Several discrete prediction time window lengths are preset, such as 5 minutes, 10 minutes, and 15 minutes, corresponding to low, medium, and high data confidence levels, respectively.

[0037] Define the confidence level interval. For example, when A value between 0.8 and 1.0 is considered high confidence; between 0.5 and 0.8 is medium confidence; and below 0.5 is low confidence. Real-time calculations are performed during control system operation. The value. If the patient's condition is stable in the initial stage and the monitoring data is reliable, If the mean arterial pressure is 0.9, a 15-minute time window is automatically selected for prediction and long-term optimization. If, at any given moment, the mean arterial pressure reading becomes abnormal due to ductus arteriosus blockage, The value drops sharply to 0.4, switching the forecast window to 5 minutes. This short-sighted adjustment allows the controller to focus more on immediate, more reliable short-term forecasts, improving system safety.

[0038] In an optional embodiment, adjusting key parameters affecting the control strategy based on the reliability of the parameters corresponding to the main treatment indicator, wherein the key parameters include the prediction time window length and the asymmetric penalty weight for deviation of the predicted trajectory from the target, includes: in the cost function, when the predicted value of the main treatment indicator is lower than the lower limit of the target range or higher than the upper limit of the target range, the penalty weight is positively correlated with the reliability of the parameters of the main treatment indicator.

[0039] Model predictive control finds the optimal control sequence by minimizing a cost function. This cost function includes a penalty term for deviations of the predicted trajectory from the target range. For the primary therapeutic endpoint, mean arterial pressure (MAP), the target is a range, such as 65 to 75 mmHg. The cost function imposes a penalty when the predicted value is below 65 or above 75. The strength of this penalty, i.e., the penalty weight, is related to the confidence level of the MAP reading. Correlation. The penalty term for mean arterial pressure in the cost function can be expressed as: .

[0040] Penalty weight here and Functions that are not based on credibility, for example ,in It is the basic weight. It is the adjustment coefficient. Assume... , .when When the value is 0.9, it indicates highly reliable data, and the penalty weight increases to 14. At this point, the controller will employ a greater control cost, such as adjusting the drug dosage, to strictly maintain the mean arterial pressure within the target range. Conversely, when... When the signal quality drops to 0.3, the penalty weight is reduced to 8. This indicates that the system's tolerance for deviations from the target has increased, avoiding overreaction to potentially inaccurate data, prioritizing the smoothness of control actions, and preventing drastic fluctuations in drug dosage caused by following noise signals, as shown in Figure 6.

[0041] In an optional embodiment, the multidimensional safety state space defined by the primary treatment indicator and the side effect precursor indicator includes: the primary treatment indicator is mean arterial pressure; the side effect precursor indicators include heart rate and stroke volume variability; the safety state space is defined as a set that satisfies all of the following conditions: the mean arterial pressure is within a preset target range; the values ​​of the heart rate and stroke volume variability are within their respective adjusted safety boundaries.

[0042] In an optional embodiment, the most clinically critical indicators are selected as the treatment target and safety constraint, respectively. Mean arterial pressure (MAP) is the primary therapeutic indicator guiding the titration of vasoactive drugs. It directly reflects the perfusion pressure of vital organs, and maintaining MAP within the normal range, such as 65 to 75 mmHg, is crucial for preventing shock and organ failure. Therefore, the core task of the control system is to precisely control the patient's future MAP trajectory within the target range by automatically adjusting the drug dosage.

[0043] To ensure treatment safety, potential drug-induced side effects must be monitored. Heart rate is a key indicator of potential side effects because vasoactive drugs, such as norepinephrine, can cause reflex bradycardia or, in some cases, tachycardia. Keeping the heart rate within a safe range, such as 50 to 110 beats per minute, mitigates these risks. Stroke volume variability is another important indicator, reflecting a patient's volume status. A high value, such as exceeding 13%, usually suggests potential hypovolemia. In such cases, vasopressors alone are ineffective and potentially harmful. Therefore, monitoring stroke volume variability and using it as a safety constraint prevents the system from erroneously increasing the vasopressor dosage when the patient requires rehydration.

[0044] This safe state space provides a set of hard constraints that the model predictive control algorithm must adhere to. In each control cycle, when searching for the optimal dosing strategy, the algorithm must not only consider how to guide the primary therapeutic indicator, i.e., mean arterial pressure, to the target range, but also ensure that all state points of the mean arterial pressure throughout the entire prediction time domain are within this safe space. This space is a multidimensional region defined by multiple inequalities. For example, suppose the state vector is... Then, at any time k in the prediction time domain, the predicted state is... Three conditions must be met simultaneously.

[0045] The first condition is the constraint of treatment goals, that is The second condition is heart rate safety constraints. boundary values and It is based on the reliability of the heart rate signal. The calculation yielded the third condition, which is a safety constraint on the variability of stroke volume. upper limit Also determined by signal reliability The controller then makes a decision. For example, the controller considers a dosing regimen that the model predicts will bring the mean arterial pressure to 70 mmHg, but will simultaneously cause the heart rate to drop to 48 beats per minute. If the confidence level of the heart rate at this point is high, and the safe lower limit of the heart rate is 52 beats per minute, then the predicted point falls outside the safe margin. Therefore, the controller determines that the dosing regimen is not feasible and discards it, instead searching for another suboptimal solution that satisfies all constraints, as shown in Figure 7S3. The first control command of the optimal dosing sequence obtained from the solution is output to the infusion pump for execution.

[0046] Specifically, the model predictive control calculates an optimal dosing sequence for a future period of time in each control cycle, such as every 10 seconds, for example, the dosing rate per minute over the next 10 minutes. This step only extracts the first control instruction from this sequence, i.e., the dosing rate corresponding to the first 10 seconds, and sends this rate to the infusion pump for execution. At the arrival of the next 10-second cycle, new physiological parameters are obtained, the confidence level is recalculated, and the entire optimization process is repeated to obtain a new dosing sequence, again executing only the first instruction. This rolling optimization method continuously responds to the patient's latest condition.

[0047] In a second embodiment, the present invention also provides an infusion pump drug dosage control system integrating clinical decision support, comprising the following modules: a calculation module, used to acquire real-time monitoring data of at least two physiological parameters, said physiological parameters including a primary therapeutic indicator and at least one precursor to side effects, and for each said physiological parameter, calculate parameter reliability based on signal fluctuation characteristics and deviation from the predicted value of a PK-PD model; a construction module, used to construct and solve a model predictive control problem to determine the optimal drug dosage based on the calculated parameter reliability of each of the said physiological parameters, wherein the model predictive control problem utilizes the parameter reliability in the following manner: during state updates In the process, the real-time monitoring data is weighted using the reliability of the parameters to correct the patient state used for model prediction; in the constraints, a multidimensional safe state space is constructed, defined by the main treatment indicators and the side effect precursor indicators, and the boundary of the safe state space is adjusted according to the reliability of the parameters corresponding to the side effect precursor indicators; in the cost function, the key parameters affecting the control strategy are adjusted according to the reliability of the parameters corresponding to the main treatment indicators, the key parameters including the prediction time window length and the asymmetric penalty weight for deviation of the prediction trajectory from the target; the output module is used to output the first control command of the optimal dosing sequence obtained by solving to the infusion pump for execution.

[0048] In this specification, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise limited, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the associated listed items.

[0049] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0050] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for controlling drug dosage using an infusion pump that integrates clinical decision support, characterized in that, Includes the following steps: Real-time monitoring data of at least two physiological parameters are acquired, including a major treatment indicator and at least one precursor indicator of side effects. For each of the physiological parameters, the reliability of the parameter is calculated based on the signal fluctuation characteristics and the deviation from the predicted value of the PK-PD model. Based on the calculated reliability of each of the physiological parameters, a model predictive control problem is constructed and solved to determine the optimal drug dosage. The model predictive control problem utilizes the parameter reliability in the following ways: in state updates, the real-time monitoring data is weighted using the parameter reliability to correct the patient state used for model prediction; in constraints, a multidimensional safe state space is constructed defined by the primary treatment indicator and the pre-side effect indicator, and the boundary of the safe state space is adjusted according to the parameter reliability corresponding to the pre-side effect indicator; in the cost function, key parameters affecting the control strategy are adjusted according to the parameter reliability corresponding to the primary treatment indicator, including the prediction time window length and the asymmetric penalty weight for deviations of the predicted trajectory from the target; the first control command of the solved optimal drug dosage sequence is output to the infusion pump for execution.

2. The method according to claim 1, characterized in that, The calculation of parameter reliability based on signal fluctuation characteristics and deviation from PK-PD model predictions includes: calculating the signal fluctuation index of the physiological parameter and the deviation index from the model predictions; and performing weighted fusion of the signal fluctuation index and the deviation index to calculate the parameter reliability.

3. The method according to claim 1, characterized in that, In the state update, the real-time monitoring data is weighted using the parameter confidence level to correct the patient state predicted by the model. This includes: using a Kalman filter for state update, wherein the real-time monitoring data is used as the observed value; and retrieving the diagonal elements of the measurement noise covariance matrix R in the Kalman filter. , set as the confidence level with respect to the corresponding physiological parameter i Inversely proportional.

4. The method according to claim 1, characterized in that, The multidimensional safety state space defined by the primary treatment indicator and the pre-side effect indicator includes: the primary treatment indicator is mean arterial pressure; the pre-side effect indicator includes heart rate and stroke volume variability; the safety state space is defined as a set that satisfies all of the following conditions: the mean arterial pressure is within a preset target range; the values ​​of the heart rate and stroke volume variability are within their respective adjusted safety boundaries.

5. The method according to claim 1, characterized in that, The step of adjusting the boundary of the safe state space based on the reliability of the parameters corresponding to the side effect precursor indicators includes: when the reliability of the parameters of a certain side effect precursor indicator decreases, shrinking the safe boundary corresponding to the reliability of the parameters towards a preset safe baseline value, wherein the amount of shrinkage is proportional to the degree of decrease in the reliability of the parameters.

6. The method according to claim 1, characterized in that, The step of adjusting key parameters affecting the control strategy based on the reliability of the parameters corresponding to the main treatment indicators includes the prediction time window length, which involves setting the prediction time window length to a corresponding discrete value based on the preset interval to which the reliability of the parameters of the main treatment indicators belongs, wherein the higher the reliability of the parameters, the longer the prediction time window length is set.

7. The method according to claim 6, characterized in that, The asymmetric penalty weight for deviation of the predicted trajectory from the target includes: in the cost function, when the predicted value of the main treatment indicator is lower than the lower limit of the target range or higher than the upper limit of the target range, the penalty weight is positively correlated with the parameter reliability of the main treatment indicator.

8. A drug dosing control system for an infusion pump integrating clinical decision support, characterized in that, It includes the following modules: a calculation module, used to acquire real-time monitoring data of at least two physiological parameters, the physiological parameters including major treatment indicators and at least one side effect precursor indicator, and to calculate the reliability of each physiological parameter based on signal fluctuation characteristics and deviation from the predicted value of the PK-PD model; A construction module is used to construct and solve a model predictive control problem to determine the optimal drug dosage based on the calculated parameter confidence of each of the physiological parameters. The model predictive control problem utilizes the parameter confidence in the following ways: in state updates, the real-time monitoring data is weighted using the parameter confidence to correct the patient state used for model prediction; in constraint conditions, a multidimensional safe state space defined by the primary treatment indicator and the side effect precursor indicator is constructed, and the boundary of the safe state space is adjusted according to the parameter confidence corresponding to the side effect precursor indicator; in the cost function, key parameters affecting the control strategy are adjusted according to the parameter confidence corresponding to the primary treatment indicator, the key parameters including the prediction time window length and the asymmetric penalty weight for deviation of the predicted trajectory from the target; and an output module is used to output the first control command of the solved optimal drug dosage sequence to the infusion pump for execution.

9. The system according to claim 8, characterized in that, The calculation of parameter reliability based on signal fluctuation characteristics and deviation from the predicted values ​​of the PK-PD model includes: calculating the signal fluctuation index of the physiological parameter and the deviation index from the predicted values ​​of the model; and performing weighted fusion based on the signal fluctuation index and the deviation index to calculate the parameter reliability, wherein the smaller the signal fluctuation and deviation, the higher the parameter reliability.

10. The system according to claim 8, characterized in that, In the state update, the real-time monitoring data is weighted using the parameter confidence level to correct the patient state predicted by the model. This includes: using a Kalman filter for state update, wherein the real-time monitoring data is used as the observed value; and retrieving the diagonal elements of the measurement noise covariance matrix R in the Kalman filter. , set as the confidence level with respect to the corresponding physiological parameter i Inversely proportional.

Images