An AI-based nursing risk prediction system

Abstract

This invention discloses an artificial intelligence-based nursing risk prediction system, comprising a nursing information acquisition module, a neighborhood-accumulated risk construction module, a nursing risk response module, a target design module, a dynamic risk increment mapping module, a model training module, and a nursing risk prediction module. This invention belongs to the field of risk prediction, specifically referring to an artificial intelligence-based nursing risk prediction system. This scheme reflects the timeliness of risk through exponential decay weighted accumulation, quantifies accumulated nursing risk through mean-neighborhood accumulation smoothing, introduces nonlinear mapping and a time-distance sensitive kernel to capture nonlinear interactions of nursing information, enhances dynamic risk capture through mean-neighborhood accumulation values ​​to improve sensitivity to instantaneous high fluctuations, filters minor nursing fluctuations through a threshold exemption function, introduces event association weighting to strengthen the prediction of accompanying acute events, and quantifies risk increments through a risk development coefficient, thereby improving the reliability of nursing risk prediction.

Description

Technical Field

[0001] This invention relates to the field of risk prediction, specifically to an artificial intelligence-based nursing risk prediction system. Background Technology

[0002] Nursing risk prediction systems collect patient data, analyze potential risk factors, and transform clinical data into risk warning information to assist medical staff in reducing the incidence of adverse events. However, general nursing risk prediction systems have several drawbacks. They ignore historical risk accumulation, fail to reflect long-term risk accumulation in patients, are sensitive to short-term nursing fluctuations, and easily filter out key instantaneous signals, leading to poor accuracy in nursing risk prediction. Furthermore, general nursing risk prediction systems treat occasional acute events as outliers, lack sensitivity to key risk signals, and exhibit delayed responses to recent risk changes, resulting in low reliability in nursing risk prediction. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an artificial intelligence-based nursing risk prediction system. Addressing the problems of general nursing risk prediction systems neglecting historical risk accumulation, failing to reflect long-term patient risk accumulation, being sensitive to short-term nursing fluctuations, and easily filtering out key instantaneous signals, thus leading to poor prediction accuracy, this solution uses exponentially decaying weighted accumulation to reflect the timeliness of risk, avoiding prediction bias caused by excessive interference from old nursing data; it quantifies accumulated nursing risk through mean neighborhood accumulation smoothing, introduces nonlinear mapping and a time-distance sensitive kernel to capture the nonlinear interaction of nursing information; and it enhances dynamic... This approach aims to capture dynamic risks and improve sensitivity to transient heart rate spikes and fluctuations, thereby enhancing the accuracy of nursing risk prediction. Addressing the issues of general nursing risk prediction systems treating occasional acute events as outliers, insufficient sensitivity to key risk signals, and delayed response to recent risk changes, leading to low reliability, this solution filters minor daily nursing fluctuations using a threshold exemption function; introduces event-related weighting to reinforce predictions for accompanying acute events, avoiding prediction bias caused by noise interference from routine nursing assessments; and quantifies risk increments using a risk development coefficient, performing dynamic risk increment mapping to further improve the reliability of nursing risk prediction.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an artificial intelligence-based nursing risk prediction system, including a nursing information collection module, a neighborhood cumulative risk construction module, a nursing risk response module, a target design module, a dynamic risk increment mapping module, a model training module, and a nursing risk prediction module;

[0005] The nursing information acquisition module obtains patient nursing information and sets the quantitative indicators as a sequence of influencing factors.

[0006] The neighborhood cumulative risk construction module generates a cumulative risk sequence and a cumulative index through exponential decay weighted accumulation, and constructs a mean neighborhood cumulative risk value.

[0007] The nursing risk response module introduces time-sensitive kernel processing to construct nursing risk response functions;

[0008] The target design module constructs a regularized risk loss target with event correlation weighting and obtains the risk development coefficient;

[0009] The dynamic risk increment mapping module generates an instant risk score prediction value based on the risk development coefficient through a kernelized response function and inverse accumulation.

[0010] The model training module uses patient nursing information as a dataset and the nursing risk response module, target design module, and dynamic risk increment mapping module as nursing risk prediction models to build the model.

[0011] The nursing risk prediction module predicts nursing risks based on the established nursing risk prediction model and real-time patient nursing information.

[0012] Furthermore, the nursing information acquisition module obtains patient nursing information; it treats the nursing risk score as a target sequence, and sets the target sequence... Represented as: ;in, This refers to the risk score during the corresponding number of nursing assessments; n is the maximum number of assessments; defining the influencing factors of patient care information. , represented as: ;in, , which is the corresponding number of nursing assessment days, category u quantitative indicators; N is the total number of indicators.

[0013] Furthermore, the neighborhood accumulation risk construction module introduces exponential decay weighted accumulation to perform exponential decay weighted accumulation, expressed as: ; ; ;in, It is a risk score sequence obtained after one accumulation; It is the attenuation factor; This is the cumulative risk score for the corresponding number of nursing assessments, where k is the number of nursing care sessions and v is the cumulative index. This corresponds to the cumulative risk indicator of category u on the nursing assessment day; the cumulative risk value is expressed as: ;in, It is the cumulative risk value of the mean neighborhood when assessing the corresponding number of times; it is more sensitive to transient high fluctuations, including brief spikes in heart rate, indicating potential acute events.

[0014] Furthermore, the nursing risk response module constructs a nursing risk response function, expressed as follows: ; Where d is the risk development coefficient; B is the weight parameter vector, corresponding to the influence intensity of the u-th type cumulative index after nonlinear mapping; d and B are unknown; T is the matrix transpose. It is a cumulative risk indicator vector; It is a nonlinear mapping that introduces a time-spacing sensitive kernel. , represented as: A and R are auxiliary vectors; and It is the kernel width parameter; and It is the time interval between the nursing assessment for A and R and the previous nursing assessment; and These are the cumulative risk indicators for categories 2 and 3 respectively, corresponding to the number of nursing assessment days.

[0015] Furthermore, the target design module combines empirical risk with structural complexity to construct a regularized target for risk loss, and introduces event correlation weighting. The objective function is expressed as: ;in, ; Where C is the penalty coefficient and M is the total number of training samples; It is a threshold exemption function, and z is an auxiliary parameter; It is the exemption threshold; , and They are the i-th patients , and ; and These are the patient's negative punishment multiplier and positive punishment multiplier, respectively. This refers to the event correlation weight. If an acute event occurs during the k-th nursing care session for the i-th patient, then... ,otherwise , It is the event correlation factor; by introducing Lagrange multipliers to construct the dual problem, it can be expressed as: ; Where u and v are the nursing frequency indices; and These are the positive penalty multipliers for the patient's u-th and v-th nursing care sessions, respectively; and These are the negative penalty multipliers for the patient's u-th and v-th nursing care sessions, respectively.

[0016] Furthermore, the dynamic risk increment mapping module, based on complementary relaxation conditions, solves for the candidate risk development coefficient for each patient. ,like ,but ;like ,but The arithmetic mean of all candidate risk development coefficients is taken as the final risk development coefficient d. The cumulative prediction of nursing risk is then calculated and expressed as: ;in, This is the cumulative risk prediction value corresponding to the number of nursing care sessions; and by inverse accumulation, the cumulative prediction value is restored to a single-step risk increment to obtain the patient's specific nursing risk score at the next moment, expressed as: ;in, It is an instant risk score prediction value.

[0017] Furthermore, the model training module uses the acquired patient care information as a dataset, and uses the nursing risk response module, target design module, and dynamic risk increment mapping module as nursing risk prediction models. The nursing risk prediction models are trained using the real-time risk score prediction values ​​obtained from the dataset to obtain the established nursing risk prediction model.

[0018] Furthermore, the nursing risk prediction module is based on the established nursing risk prediction model, collects patient nursing information in real time and inputs it into the nursing risk prediction model, and uses the obtained real-time risk score prediction value as the prediction result.

[0019] The beneficial effects achieved by the present invention using the above solution are as follows:

[0020] (1) In view of the problems that general nursing risk prediction systems ignore historical risk accumulation, cannot reflect the long-term risk accumulation of patients, are sensitive to short-term nursing fluctuations, and easily filter out key instantaneous signals, thus leading to poor accuracy of nursing risk prediction, this solution reflects the timeliness of risk through exponential decay weighted accumulation, avoiding prediction bias caused by excessive interference from old nursing data; it quantifies accumulated nursing risk through mean neighborhood accumulation smoothing, introduces nonlinear mapping and time distance sensitive kernel to capture nonlinear interaction of nursing information; it enhances dynamic risk capture through mean neighborhood accumulation value, improves sensitivity to transient heart rate spikes and instantaneous high fluctuations; and thus improves the accuracy of nursing risk prediction.

[0021] (2) In view of the problems that general nursing risk prediction systems treat occasional acute events as outliers, are not sensitive enough to key risk signals, and are slow to respond to recent risk changes, resulting in low reliability of nursing risk prediction, this solution filters out minor fluctuations in daily nursing care based on a threshold exemption function; introduces event association weighting to strengthen the prediction of accompanying acute events and avoid prediction bias caused by noise interference from ordinary nursing assessments; and quantifies risk increments through risk development coefficients and performs dynamic risk increment mapping to improve the reliability of nursing risk prediction. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an artificial intelligence-based nursing risk prediction system provided by the present invention.

[0023] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0025] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0026] Example 1, see Figure 1 The present invention provides an artificial intelligence-based nursing risk prediction system, including a nursing information acquisition module, a neighborhood cumulative risk construction module, a nursing risk response module, a target design module, a dynamic risk increment mapping module, a model training module, and a nursing risk prediction module.

[0027] The nursing information acquisition module obtains patient nursing information, sets quantitative indicators as an influencing factor sequence, and sends the data to the neighborhood cumulative risk construction module.

[0028] The neighborhood cumulative risk construction module generates a cumulative risk sequence and a cumulative index through exponential decay weighted accumulation, constructs a mean neighborhood cumulative risk value, and sends the data to the nursing risk response module.

[0029] The nursing risk response module incorporates time-sensitive kernel processing to construct a nursing risk response function and sends the data to the target design module.

[0030] The target design module constructs a regularized risk loss target with event correlation weighting and obtains the risk development coefficient; and sends the data to the dynamic risk increment mapping module.

[0031] The dynamic risk increment mapping module generates an instantaneous risk score prediction value based on the risk development coefficient through a kernelized response function and inverse accumulation; and sends the data to the model training module.

[0032] The model training module uses patient nursing information as a dataset, and uses the nursing risk response module, target design module, and dynamic risk increment mapping module as nursing risk prediction models to build the model; and sends the data to the nursing risk prediction module.

[0033] The nursing risk prediction module predicts nursing risks based on the established nursing risk prediction model and real-time patient nursing information.

[0034] Example 2, see Figure 1 This embodiment is based on the above embodiment. The nursing information collection module acquires patient nursing information; it treats the nursing risk score as a target sequence, and uses the historically accumulated comprehensive risk assessment value to reflect the patient's potential risk level; let the target sequence... Represented as: ;in, This refers to the risk score corresponding to the number of nursing assessments, with the actual label attached, on a percentage basis; n is the maximum number of assessments, i.e., the total number of historical nursing assessments considered; defining the influencing factors of patient care information. , represented as: ;in, This refers to the u-th category of quantitative indicators on the corresponding nursing assessment day; N is the total number of indicators; the quantitative indicators of patient nursing information include: average blood pressure, pain score, number of steps taken, average body temperature, serum albumin level, medication completion rate, and number of nighttime awakenings; the quantitative indicators need to be processed by feature engineering.

[0035] Example 3, see Figure 1 This embodiment is based on the above embodiment. The neighborhood cumulative risk construction module introduces exponential decay weighted accumulation. Nursing risks often accumulate over time, including prolonged bed rest-pressure ulcers and long-term low protein levels-infection. Accumulation smooths out random fluctuations in the scoring sequence and factor sequence. However, the historical accumulation of nursing risks is not equally weighted. Unlike long-term risks such as prolonged bed rest-pressure ulcers, more recent falls and low protein levels will have a more direct impact on the current risk level. Therefore, exponential decay weighted accumulation is performed, expressed as: ; ; ;in, It is a risk score sequence obtained after one accumulation; It is the attenuation factor; This is the cumulative risk score for the corresponding number of nursing assessments, where k is the number of nursing care sessions and v is the cumulative index. This corresponds to the cumulative risk index of category u on the nursing assessment day; and by using the mean of the time points before and after, an equivalent smooth node is constructed to enhance the fitting accuracy of the time series evolution, facilitating the physical interpretation of the subsequent risk development coefficient d—that is, the average increment of risk between each assessment. The cumulative risk value is expressed as: ;in, It is the cumulative risk value of the mean neighborhood when assessing the corresponding number of times; it is more sensitive to transient high fluctuations, including brief spikes in heart rate, indicating potential acute events.

[0036] Example 4, see Figure 1 This embodiment is based on the above embodiment. The nursing risk response module linearly combines the risk increment and the original risk score on the left, and linearly weights the multidimensional factors after nonlinear mapping φ on the right. It can simultaneously capture the complex nonlinear interaction of each factor on the risk. The nursing risk response function is expressed as: ; Where d is the risk development coefficient; B is the weight parameter vector, corresponding to the influence intensity of the u-th type cumulative index after nonlinear mapping; d and B are unknown; T is the matrix transpose. It is a cumulative risk indicator vector; It is a nonlinear mapping that introduces a time-spacing sensitive kernel. , represented as: A and R are auxiliary vectors; and It is the kernel width parameter; and It represents the time interval between the nursing assessment at A and R and the previous nursing assessment; it is used to handle nursing sampling at unequal intervals and is more sensitive to emergencies. and These are the cumulative risk indicators for categories 2 and 3 respectively, corresponding to the number of nursing assessment days.

[0037] By performing the above operations, this solution addresses the problems of general nursing risk prediction systems, such as ignoring historical risk accumulation, failing to reflect long-term patient risk accumulation, being sensitive to short-term nursing fluctuations, and easily filtering out key instantaneous signals, thus leading to poor accuracy in nursing risk prediction. This solution uses exponentially decaying weighted accumulation to reflect the timeliness of risk, avoiding prediction bias caused by excessive interference from old nursing data; it quantifies accumulated nursing risk through mean neighborhood accumulation smoothing, introduces nonlinear mapping and a time-distance sensitive kernel to capture nonlinear interactions of nursing information; and it enhances dynamic risk capture through mean neighborhood accumulation, improving sensitivity to transient high fluctuations in heart rate; thereby improving the accuracy of nursing risk prediction.

[0038] Example 5, see Figure 1This embodiment is based on the above embodiment. The target design module constructs a regularized target for risk loss by combining empirical risk with structural complexity. It does not penalize prediction errors below a threshold, but only penalizes abnormal differences, thereby enhancing the model's robustness to incidental nursing events, including a single accidental fall leading to an exceptionally high score. Event association weighting is introduced. In nursing scenarios, different assessment points may be accompanied by known acute events, and prediction errors at these moments should be penalized more heavily. The objective function is expressed as: ;in, ; Where C is the penalty coefficient and M is the total number of training samples; It is a threshold exemption function, and z is an auxiliary parameter; It is the exemption threshold; , and They are the i-th patients , and ; and These are the patient's negative punishment multiplier and positive punishment multiplier, respectively. This refers to the event correlation weight. If an acute event occurs during the k-th nursing care session for the i-th patient, then... ,otherwise , It is the event correlation factor; by introducing Lagrange multipliers to construct the dual problem, it can be expressed as: ; Where u and v are the nursing frequency indices; and These are the positive penalty multipliers for the patient's u-th and v-th nursing care sessions, respectively; and These are the negative penalty multipliers for the patient's u-th and v-th nursing care sessions, respectively; the kernel function flexibly captures the nonlinear interactions between multidimensional clinical indicators, improving the ability to identify complex risk signal patterns; and the Lagrange multipliers and support vector indices are obtained by calling a general quadratic programming solver.

[0039] Example 6, see Figure 1 This embodiment is based on the above embodiment. The dynamic risk increment mapping module, based on complementary relaxation conditions, solves for the candidate risk development coefficient for each patient. ,like ,but ;like ,but The arithmetic mean of all candidate risk development coefficients is taken as the final risk development coefficient d. The cumulative prediction of nursing risk is then calculated and expressed as: ;in, This is the cumulative risk prediction value corresponding to the number of nursing care sessions. Through a kernelized response function, the cumulative nursing risk value at future time points is represented as a weighted kernel product of historical support vectors and new factor vectors, possessing both nonlinear fitting and time decay characteristics. It incorporates decay memory, where the influence of older measurements on the current prediction decreases, aligning with the experience that proximal nursing procedures determine near-term risk. Finally, through inverse accumulation generation, the cumulative prediction value is restored to a single-step risk increment, yielding the patient's specific nursing risk score for the next time point, expressed as: ;in, It is an instant risk score prediction value.

[0040] By performing the above operations, this solution addresses the problems of general nursing risk prediction systems, such as treating occasional acute events as outliers, insufficient sensitivity to key risk signals, and delayed response to recent risk changes, leading to low reliability in nursing risk prediction. Specifically, it filters out minor fluctuations in daily nursing care based on a threshold exemption function; introduces event association weighting to strengthen predictions for accompanying acute events, avoiding prediction bias caused by noise interference from routine nursing assessments; and quantifies risk increments through a risk development coefficient, performing dynamic risk increment mapping to improve the reliability of nursing risk prediction.

[0041] Example 7, see Figure 1 This embodiment is based on the above embodiment. The model training module uses the acquired patient care information as a dataset, and uses the nursing risk response module, target design module and dynamic risk increment mapping module as nursing risk prediction models. The nursing risk prediction model is trained by the real-time risk score prediction value obtained from the dataset, and a complete nursing risk prediction model is obtained.

[0042] Example 8, see Figure 1 This embodiment is based on the above embodiment. The nursing risk prediction module is based on the established nursing risk prediction model. It collects patient nursing information in real time and inputs it into the nursing risk prediction model. The obtained instant risk score prediction value is used as the prediction result. A risk score threshold is set. If the instant risk score prediction value is higher than the risk score threshold, an early warning is issued to the management personnel.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0044] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A nursing risk prediction system based on artificial intelligence, characterized in that: The system includes a nursing information collection module, a neighborhood cumulative risk construction module, a nursing risk response module, a goal design module, a dynamic risk increment mapping module, a model training module, and a nursing risk prediction module; The nursing information acquisition module obtains patient nursing information and sets the quantitative indicators as a sequence of influencing factors. The neighborhood cumulative risk construction module generates a cumulative risk sequence and a cumulative index through exponential decay weighted accumulation, and constructs a mean neighborhood cumulative risk value. The nursing risk response module introduces time-sensitive kernel processing to construct nursing risk response functions; The target design module constructs a regularized risk loss target with event correlation weighting and obtains the risk development coefficient; The dynamic risk increment mapping module generates an instant risk score prediction value based on the risk development coefficient through a kernelized response function and inverse accumulation. The model training module uses patient nursing information as a dataset and the nursing risk response module, target design module, and dynamic risk increment mapping module as nursing risk prediction models to build the model. The nursing risk prediction module predicts nursing risks based on real-time patient nursing information using the established nursing risk prediction model. The nursing risk response module constructs a nursing risk response function, expressed as follows: ; Where d is the risk development coefficient; B is the weight parameter vector, corresponding to the influence intensity of the u-th type cumulative index after nonlinear mapping; d and B are unknown; T is the matrix transpose. It is a cumulative risk indicator vector; It is a nonlinear mapping that introduces a time-spacing sensitive kernel. , is represented as: A and R are auxiliary vectors; and It is the kernel width parameter; and It is the time interval between the nursing assessment for A and R and the previous nursing assessment; and These are the cumulative risk indicators for Category 2 and Category N corresponding to the number of nursing assessment days; This refers to the risk score during the corresponding number of nursing assessments; It is the cumulative risk value of the mean neighborhood when assessing the corresponding number of times, and k is the number of times of care.

2. The nursing risk prediction system based on artificial intelligence according to claim 1, characterized in that: The nursing information acquisition module obtains patient nursing information; it treats the nursing risk score as a target sequence, and sets the target sequence... Represented as: Where n is the maximum number of assessments; define the influencing factors of patient care information. , is represented as: ;in, , which is the corresponding number of nursing assessment days, category u quantitative indicators; N is the total number of indicators.

3. The nursing risk prediction system based on artificial intelligence according to claim 2, characterized in that: The neighborhood accumulation risk construction module introduces exponential decay weighted accumulation, which is expressed as: ; ; ;in, It is a risk score sequence obtained after one accumulation; It is the attenuation factor; It represents the cumulative risk score for the corresponding number of nursing assessments; v is the cumulative index; This corresponds to the cumulative risk indicator of category u on the nursing assessment day; the cumulative risk value is expressed as: .

4. The nursing risk prediction system based on artificial intelligence according to claim 3, characterized in that: The target design module combines empirical risk with structural complexity to construct a regularized target for risk loss, and introduces event correlation weighting. The objective function is expressed as: ;in, ; Where C is the penalty coefficient and M is the total number of training samples; It is a threshold exemption function, and z is an auxiliary parameter; It is the exemption threshold; , and They are the i-th patients , and ; and These are the patient's negative punishment multiplier and positive punishment multiplier, respectively. This refers to the event correlation weight. If an acute event occurs during the k-th nursing care session for the i-th patient, then... ,otherwise , It is the event correlation factor; by introducing Lagrange multipliers to construct the dual problem, it can be expressed as: ; Where u and v are the nursing frequency indices; and These are the positive penalty multipliers for the patient's u-th and v-th nursing care sessions, respectively; and These are the negative penalty multipliers for the patient's u-th and v-th nursing care sessions, respectively.

5. The nursing risk prediction system based on artificial intelligence according to claim 4, characterized in that: The dynamic risk increment mapping module, based on complementary relaxation conditions, solves for the candidate risk development coefficient for each patient. ,like ,but ;like ,but The arithmetic mean of all candidate risk development coefficients is taken as the final risk development coefficient d. The cumulative prediction of nursing risk is then calculated and expressed as: ;in, This is the cumulative risk prediction value corresponding to the number of nursing care sessions; and by inverse accumulation, the cumulative prediction value is restored to a single-step risk increment to obtain the patient's specific nursing risk score at the next moment, expressed as: ;in, It is an instant risk score prediction value.

6. The nursing risk prediction system based on artificial intelligence according to claim 5, characterized in that: The model training module uses the acquired patient care information as a dataset, and uses the nursing risk response module, target design module, and dynamic risk increment mapping module as nursing risk prediction models. The nursing risk prediction models are trained using the real-time risk score prediction values ​​obtained from the dataset to obtain a completed nursing risk prediction model.

7. The nursing risk prediction system based on artificial intelligence according to claim 6, characterized in that: The nursing risk prediction module is based on the established nursing risk prediction model. It collects patient nursing information in real time and inputs it into the nursing risk prediction model, and uses the obtained real-time risk score prediction value as the prediction result.

Images