Neurosurgical nursing evaluation method based on clinical early warning

Abstract

The invention belongs to the technical field of medical care, and particularly relates to a neurosurgery nursing evaluation method based on clinical early warning. The method comprises the following steps: acquiring historical vital sign data of a monitored patient, detecting abnormal data points by adopting an isolated forest algorithm, smoothing time sequence data through local weighted regression, and eliminating noise data; filling data missing values by using an interpolation method; segmenting the time sequence data by using a sliding window method, and extracting trend and fluctuation feature data; aiming at each type of vital sign data, respectively constructing a time fluctuation curve based on fluctuation characteristic data, and calculating a fluctuation index of the time fluctuation curve before a critical point by taking a postoperative stationary phase as the critical point; and on the basis of the fluctuation index, comparison with the threshold value output by the machine learning model is carried out, abnormal signal identification is carried out, and the clinical early warning accuracy is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical care, and in particular relates to a neurosurgery nursing assessment method based on clinical early warning. Background Art

[0002] Neurosurgical conditions often involve complex surgical interventions on the brain or spinal cord, which can lead to numerous postoperative complications, including increased intracranial pressure, inadequate cerebral perfusion, and abnormal cerebral oxygen saturation. The rapid and complex nature of postoperative disease progression makes it difficult for traditional nursing methods (which rely on manual experience and static scoring mechanisms) to achieve accurate monitoring and rapid response. This deficiency can lead to failure to detect deterioration in a timely manner, increasing the risk of complications. Furthermore, the medical and nursing pressure in neurosurgery wards is considerable, comparable to that of other departments, and neurosurgical intensive care units place an even greater emphasis on timeliness.

[0003] However, traditional nursing models rely on manual judgment and experience, making it difficult to accurately predict and dynamically adjust complex conditions. This can lead to delays in timely intervention when conditions worsen, compromising quality of care. Therefore, there is an urgent need for intelligent nursing assessment methods that integrate data-driven analysis with real-time clinical warnings.

[0004] Currently available nursing assessment systems mostly use static scoring mechanisms, which are difficult to adapt to the dynamic changes in patients' conditions, fail to fully utilize multidimensional data characteristics, and have significant deficiencies in real-time and accuracy. Summary of the Invention

[0005] To address these technical issues, the present invention provides a neurosurgery nursing assessment method based on clinical early warning. By incorporating dynamic vital sign monitoring, multi-scale condition change analysis, and an early warning mechanism, this method provides more accurate and comprehensive nursing support for neurosurgery patients. The method includes steps such as data acquisition, fluctuation feature extraction, output of a fluctuation index, real-time anomaly detection, multi-level clinical early warning, and adjustment of nursing measures.

[0006] Specifically, the scheme of the present invention is as follows: A neurosurgery nursing assessment method based on clinical early warning, including: S1. Obtain the patient's historical vital signs data, including but not limited to: body temperature, blood pressure, respiratory rate, pulse, blood oxygen saturation, and neurophysiological parameters; Wherein, the neurophysiological parameters include intracranial pressure and cerebral oxygen saturation.

[0007] S2. Data cleaning and standardization are performed on the collected raw data: the isolation forest algorithm is used to detect abnormal data points, and the time series data is smoothed by locally weighted regression (LOWESS) to remove noise data; missing data values ​​are filled using interpolation; the time series data is segmented using the sliding window method to extract trend and fluctuation characteristics; different data types are uniformly encoded into vector representations; S3. For each type of vital sign data, a time fluctuation curve is constructed based on the fluctuation characteristics. The postoperative stable period is used as the critical point, and the fluctuation index of the time fluctuation curve before the critical point is calculated. The postoperative stable period is determined based on the different surgical procedures, or a GSC score ≥ 15 is used as the default postoperative stable period.

[0008] The time fluctuation curve fluctuation index formula is as follows: ,in: , is the ordinate value of the first peak in the time fluctuation curve; , is the ordinate value of the last peak in the time fluctuation curve; , is the difference in ordinates between adjacent peaks and troughs in the time fluctuation curve; S is the number of troughs and peaks in the time fluctuation curve.

[0009] When the monitored patient is in a coma or the cerebral oxygen saturation is lower than 65%, the cerebral oxygen saturation fluctuation index G rSO2 The formula is as follows: rSO2(t): cerebral oxygen saturation at time t; : The average value of brain oxygen saturation in time period T.

[0010] G i Indicates the The volatility index of the parameter; is the weight, reflecting the contribution of each parameter to the overall fluctuation; is the total number of parameters selected.

[0011] S4. Real-time collection of the patient's vital signs data, and dynamic detection of the data based on the fluctuation index G. The detection process includes: S41. Multi-scale feature analysis: Extract multi-scale time series features using convolutional neural networks (CNNs) and classify high-frequency anomalies and low-frequency trends to train machine learning models using classified data. S42, anomaly detection model: uses machine learning models to identify abnormal signals and update anomaly thresholds in real time; S43. Dynamically adjust the warning threshold according to the conditional probability.

[0012] Threshold judgment: Compare real-time data with the set threshold. If the threshold is exceeded, a clinical warning will be triggered to remind you to strengthen nursing intervention.

[0013] The beneficial effects of the present invention are: Based on the special nursing requirements of neurosurgery, the evaluation method of the present invention reduces the requirements for the original data dimension and further improves the warning stability. It smoothes the time series data through local weighted regression, segments the time series data using the sliding window method, extracts the fluctuation characteristics, and then obtains the comprehensive fluctuation index, which significantly reduces the data dimension and computational complexity; extracts multi-scale time series features through convolutional neural networks, and classifies high-frequency anomalies and low-frequency trends, thereby further training the machine learning model with high-frequency anomalies and low-frequency trends to obtain a dynamically adjustable warning threshold, which significantly improves the accuracy of clinical warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0015] Figure 1 The figure is a flowchart of the steps of a neurosurgery nursing assessment method based on clinical early warning in an embodiment. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the present application, the present application will be further elaborated in conjunction with specific implementation methods.

[0017] Example 1 A neurosurgery nursing assessment method based on clinical early warning, including: S1. Obtain the patient's historical vital signs data, including body temperature, blood pressure, respiratory rate, pulse, blood oxygen saturation, intracranial pressure, and brain oxygen saturation; S2. Use the isolation forest algorithm to detect abnormal data points and smooth the time series data through local weighted regression to eliminate noise data. The LOWESS smoothing window size is adapted to different parameter characteristics by calculating the standard deviation and volatility of the time series according to the following standards: Table 1 LOWESS smoothing window size standard parameter characteristic Bandwidth selection body temperature Stable, slow changes Large bandwidth (30% of data length) blood pressure There are cyclical fluctuations Medium bandwidth (20% of data length) pulse High frequency changes, high noise Small bandwidth (10% of data length) Brain oxygen saturation Susceptible to external interference and high-frequency mutation Dynamic bandwidth, combined with volatility adjustment Intracranial pressure, respiration The two parameters are highly correlated Combined adjustment Historical data is divided into training and validation sets, and the sensitivity results are used to feedback the fluctuation of the specific bandwidth value (limited to 10%).

[0018] Use linear interpolation to fill missing data values; use the sliding window method to segment time series data and extract trend and fluctuation characteristics; S3. For each type of vital sign data, a time fluctuation curve is constructed based on the fluctuation characteristics to compensate for the defect of linear interpolation in capturing complex changing trends and further improve the real-time detection effect. The postoperative stable period is used as the critical point, and the fluctuation index of the time fluctuation curve before the critical point is calculated; The postoperative stable period is determined based on the different surgical procedures, or a GSC score ≥ 12 is used as the default postoperative stable period.

[0019] The time fluctuation curve fluctuation index formula is as follows: ,in: , is the ordinate value of the first peak in the time fluctuation curve; , is the ordinate value of the last peak in the time fluctuation curve; , is the difference in ordinates between adjacent peaks and troughs in the time fluctuation curve; S is the number of troughs and peaks in the time fluctuation curve.

[0020] Establish a dynamic weight mechanism: automatically adjust according to the patient's specific postoperative condition, or use a prognostic machine learning model to output a weight index: normalize historical vital signs data, use random forest to capture nonlinear relationships, train the model, update the weights in real time through a sliding window mechanism, use cross-validation to evaluate the generalization performance of the model, and test the accuracy of weight learning on an independent test set.

[0021] After establishing the weight mechanism, we get G 综合 result: G i Indicates the The volatility index of the parameter; is the weight, reflecting the contribution of each parameter to the overall fluctuation; including w 体温 、w 血压 、w 呼吸 、w 脉搏 、w 血氧 、w 脑氧 、w 颅内压 , .

[0022] 𝑁 is the total number of selected parameters.

[0023] Among them, for patients in coma or with low brain oxygen levels measured for the first time, the brain oxygen saturation fluctuation index G rSO2 The formula is as follows: rSO2(t): cerebral oxygen saturation at time t; : The average value of brain oxygen saturation in time period T.

[0024] S4. Real-time collection of the patient's vital signs data, and dynamic detection of the data based on the fluctuation index G. The detection process includes: S41. Multi-scale feature analysis: Extract multi-scale time series features through convolutional neural networks, classify high-frequency anomalies and low-frequency trends, and use the classified data to train machine learning models; S42, anomaly detection model: uses a machine learning model to identify abnormal signals and updates the anomaly threshold in real time, denoted as G threshold; S43. Dynamically adjust the warning threshold based on conditional probability: Threshold judgment: Compare real-time data with the set threshold. If the threshold is exceeded, a clinical warning will be triggered to remind you to strengthen nursing intervention: If G 综合 ≤G 阈值 : The condition is stable; If G 综合 >G 阈值 : The condition is abnormal and an early warning is required.

[0025] Example 2 The use of a multi-level threshold strategy can effectively reduce the risk of false alarms or missed alarms caused by a single threshold. This embodiment can significantly improve the sensitivity and reliability of the evaluation method by combining conditional probability, historical data distribution, and real-time data to dynamically adjust the threshold, as follows: Statistical indicators determine T1 (low risk threshold) and T2 (high risk threshold): k1 and k2 are hyperparameters output by a random forest model with dynamic thresholds. Specifically, the values ​​of the k1 and k2 hyperparameters are determined by performing simulated early warnings based on historical vital sign data and verifying the false alarm and false negative rates. For situations with limited historical data, k1 and k2 can be determined by combining data characteristics, clinical expertise, and nursing experience. The empirically recommended values ​​are: k1 = 1, k2 = 2.

[0026] G 综合 <T1, Low risk threshold: Triggered when there are slight fluctuations, reminding to observe, no immediate intervention required.

[0027] T1 ≤ G 综合 <T2, Medium risk threshold: When significant fluctuations occur, trigger an early warning and recommend strengthening care.

[0028] G 综合 ≥ T2, High risk threshold: Seriously abnormal, immediately alarm and initiate emergency care intervention measures.

[0029] Example 3 The patient is a middle-aged male (45 years old) in a coma due to post-operative traumatic brain injury. When admitted to the hospital, the GSC score was 9, and the intracranial pressure reached 25 mmHg. He immediately entered the intensive care unit (ICU) after the operation. The nursing goal is to detect the risk of abnormal intracranial pressure fluctuations, cerebral oxygen deficiency or systemic inflammatory response as early as possible through real-time monitoring of vital signs.

[0030] Collected parameters: The vital sign data of the patient after the operation. Taking the initial observation data as an example, body temperature: 37.6 °C (the initial temperature after the operation is slightly higher, which may indicate an inflammatory response), blood pressure: 130 / 85 mmHg, respiratory rate: 16 times / minute, pulse: 82 times / minute, blood oxygen saturation (SpO2): 94%, cerebral oxygen saturation (rSO2): 57% (lower than the normal value), intracranial pressure (ICP): continuously higher than the normal value (22 - 25 mmHg) after the operation, then the cerebral oxygen saturation G rSO2 Applicable to: Data characteristics: Collect the multi-day data of the patient before and after the operation, and mark the time point when the GSC ≥ 12 at the 24th hour after the operation as the initial reference value of the stable period after the operation.

[0031] [[ID=三十一]]Use the Isolation Forest algorithm to detect abnormal data points in vital signs. There was transient noise in the cerebral oxygen saturation data at the 6th hour after the operation, which has been removed. Use the locally weighted regression method to smooth the time series. Data completion: Use the interpolation method to fill in the missing collection points in the respiratory rate, and segment by a sliding window to extract the trends and fluctuation characteristics of parameters such as body temperature and blood pressure; Based on the above formula output, G rSO2 = 1.4, w 体温 = 0.15, w 血压 = 0.2, w 呼吸 = 0.1, w 脉搏 = 0.1, w 血氧 = 0.15, w 脑氧 = 0.15, w 颅内压=0.2, based on the time fluctuation curve fluctuation index formula in Example 1, output G 综合 It is 1.73.

[0032] The patient's real-time data collection and dynamic detection are carried out, and convolutional neural networks are used to extract the characteristics of multi-parameter time series such as body temperature and brain oxygen, separate high-frequency abnormalities (sharp drop in brain oxygen during surgery) and low-frequency trends (slow increase in intracranial pressure after surgery), and a machine learning model is used to evaluate the comprehensive fluctuation index G in real time. 综合 , the initial G threshold = 1.5, which is dynamically adjusted according to the machine learning model.

[0033] First clinical warning trigger: 36 hours after surgery, fluctuation index G 综合 >2.2, where cerebral oxygen saturation G rSO2 >2.1, indicating the risk of cerebral hypoxia and increased risk.

[0034] Verification: After 3 hours, brain oxygen and blood pressure dropped, triggering the instrument prompt. Further testing of arterial carbon dioxide partial pressure and blood lactate levels confirmed an increased risk of infection and worsening of the condition, indicating that the clinical warning was accurate.

[0035] Adjustment of nursing plan: increase oxygen inhalation, adjust oxygen concentration to 50%, adjust intracranial pressure drugs (such as mannitol), strengthen dehydration treatment, and increase the frequency of cerebral oxygen saturation and intracranial pressure testing.

Claims

1. A neurosurgery nursing assessment method based on clinical early warning, characterized in that: include: S1. Obtaining historical vital sign data of the monitored patient, wherein the vital sign data includes body temperature, blood pressure, respiration, pulse, blood oxygen saturation and neurophysiological parameters; S2, using the isolation forest algorithm to detect abnormal data points, smoothing time series data through local weighted regression to remove noise data; using interpolation to fill missing data values; using the sliding window method to segment time series data and extract trend and fluctuation feature data; S3. For each type of vital sign data, a time fluctuation curve is constructed based on the fluctuation characteristic data, and the postoperative stable period is taken as the critical point, and the fluctuation index of the time fluctuation curve before the critical point is calculated; The time fluctuation curve fluctuation index formula is as follows: ,in: , is the ordinate value of the first peak in the time fluctuation curve; , is the ordinate value of the last peak in the time fluctuation curve; , is the difference in ordinates between adjacent peaks and troughs in the time fluctuation curve; S is the number of troughs and peaks in the time fluctuation curve; S4. Real-time collection of the patient's vital signs data, and dynamic detection of the data based on the fluctuation index G. The detection process includes: S41. Multi-scale feature analysis: Extract multi-scale time series features through convolutional neural networks, classify high-frequency anomalies and low-frequency trends, and train machine learning models with classified data; S42, anomaly detection model: use machine learning models to identify abnormal signals and update anomaly thresholds in real time; S43. Dynamically adjust the warning threshold according to conditional probability. Threshold judgment: compare the real-time data with the set threshold. If it exceeds the threshold, a clinical warning is triggered to remind you to strengthen nursing intervention.

2. A neurosurgery nursing assessment method based on clinical early warning as claimed in claim 1, characterized in that: The neurophysiological parameters include intracranial pressure and cerebral oxygen saturation.

3. A neurosurgery nursing assessment method based on clinical early warning as described in claim 2, characterized in that: When the monitored patient is in a coma or the cerebral oxygen saturation is lower than 65%, the fluctuation index formula of the cerebral oxygen saturation time fluctuation curve is as follows: rSO2(t): cerebral oxygen saturation at time t; : The average value of brain oxygen saturation in time period T.

4. A neurosurgery nursing assessment method based on clinical early warning as claimed in claim 1, characterized in that: In S41, the multi-scale feature analysis: extract multi-scale time series features through a convolutional neural network, and classify high-frequency anomalies and low-frequency trends. The high-frequency anomalies are abnormal data during surgery and changes of more than 3% in postoperative blood oxygen saturation and intracranial pressure within 1 second.

5. A neurosurgery nursing assessment method based on clinical early warning as claimed in claim 1, characterized in that: Including S5, multi-level thresholds are used to reduce the risk of false alarms or missed alarms caused by a single threshold, and the thresholds are adjusted dynamically based on historical data distribution and real-time data: The low-risk threshold T1 and high-risk threshold T2 are determined by statistical indicators: Among them, k1 and k2 are the outputs of the random forest model trained by the hyperparameters through false / missing data. is the mean value of G, is the standard deviation of G; G 综合 <T1, low risk threshold: Triggered when there are minor fluctuations, reminding to observe and no immediate intervention is required; T1 ≤ G 综合 <T2, medium risk threshold: significant fluctuations occur, triggering an alarm and suggesting enhanced care; G 综合 ≥T2, high-risk threshold: severe abnormality, immediate alarm and initiation of emergency nursing interventions.

6. A neurosurgery nursing assessment system based on clinical early warning, comprising: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, a neurosurgery nursing assessment method based on clinical early warning as described in any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, a neurosurgery nursing assessment method based on clinical early warning is implemented as described in any one of claims 1 to 5.

Images