TWI704509B - Patient accident risk evaluation system and method - Google Patents

Abstract

A patient accident risk evaluation system and a method thereof are provided. The method includes steps of: using a patient safety reporting platform to store information of incident events occurring to patients; using a medical record database to store medical record of the patients admitted to a hospital; obtaining the information of the incident events and the medical record having features on an admission evaluation form; determining correlation between the features and the incident event; removing the features, each of which the correlation with the incident event is lower than a low correlation threshold or a high correlation threshold; using machine learning to analyze the feature and the incident event that have highest correlation to each other to establish a most suitable model; and evaluating risk of the accident event occurring to new patients based on the most suitable model.

Description

Patient accident risk assessment system and method

The invention relates to an assessment system, in particular to a patient accident risk assessment system and method.

Patient safety is the foundation of healthcare services. In its 1999 book "To Err is Human: Building a Safer Health System" published by the Institute of Medicine (IOM), the Institute of Medicine (IOM) revealed that the United States is caused by preventable medical errors (Preventable Medical Errors) on average every year. The number of deaths is estimated to be between 44,000 and 98,000, which is much higher than the number of accidental deaths caused by traffic accidents each year. The financial losses caused by them are approximately US$17-29 billion.

In order to reduce the harm caused by medical errors to patients, governments of various countries have set up relevant agencies to implement the promotion of patient safety. For example, the British government established the National Patient Safety Agency (NPSA) in 2001, and the Australian government established the Australian Council for Safety and Quality in HealthCare.

In Taiwan, the Department of Health formally established the "Patient Safety Committee" in February 2003 to kick off the advancement of patient safety affairs and coordinate and promote patient safety related work. In 2004, the "Patient Safety" was defined by the Hospital Evaluation and Medical Quality Policy Association (Medical Council). This definition is "avoidance, prevention, and improvement of adverse results or damage caused by the health care process Measures. These bad results or injuries include errors, deviations and accidents." The goals of the Medical Policy Council for the hospital's medical quality and patient safety during the 105-108 years are: 1. Improve effective communication between medical and nursing staff, 2. Implement patient safety incident management, 3. Improve surgical safety, 4. Prevent patients from falling And reduce the degree of injury, 5. Improve drug safety, 6. Implement infection control, 7. Improve pipeline safety, 8. Encourage patients and their families to participate in patient safety work.

For example, from 2005 to 2016, the number of falls in medical institutions in Taiwan has always been the top three accidents in medical institutions. Research by Baker et al. showed that the probability of accidents in hospitalized patients is approximately 2.9%-16.6%. In addition, Alexander et al. found that patients who had a fall would spend 5.3% more in hospital expenses than other patients. In Taiwan, patients with a fall will be hospitalized 6.4 days longer than those without a fall, and cost 23,339.2 Taiwan dollars more. The injuries of inpatients caused by falls increase the consumption of medical resources and the investment in nursing care. Therefore, actively preventing the occurrence of accidents such as patient falls is very important for reducing medical expenses and improving the quality of medical services.

Patient safety has always been an important issue for improving medical quality. How to screen high-risk groups that are prone to accidents is the first step to prevent accidents. In order to prevent the possibility of non-occurrence based on the conditions that have occurred, the present invention provides a method and system for patient accident risk assessment.

The present invention provides a patient accident risk assessment method, which includes the following steps: using a patient safety notification platform to store an accident data of each accident that occurs for each patient; using a medical record database to store the medical records of each patient to the hospital for consultation Data; establish an admission evaluation form, which has multiple characteristics; obtain multiple medical records with multiple characteristics and multiple accident data of pre-evaluated accidents; based on multiple medical records and multiple accident data, Compare the relevance of each feature with the unexpected event to be assessed for feature screening; when performing feature screening, exclude features whose relevance is lower than the threshold of too low relevance, and exclude those whose relevance is higher than the threshold of too high relevance Features; using machine learning, from the multiple features retained after feature screening, analyze the features that have the highest relevance to the unexpected event to be evaluated to establish an optimal model; and based on the optimal model, fill in according to the newly admitted patients The multiple features of the admission assessment form assess the risk of accidents for newly admitted patients.

In one embodiment, the steps performed by using machine learning in the method for assessing patient accident risk include: using machine learning, from multiple features retained after feature screening, it is analyzed that the accidental event to be evaluated has a relative comparison. Multiple features with high relevance can be used to establish a training model; and machine learning can be used to test out the features in the training model that have the highest relevance to the unexpected event to be evaluated to establish an optimal model.

In one embodiment, the patient accident risk assessment method further includes the following steps: setting an assessment time range; and extracting multiple medical records within the assessment time range and multiple accident data of pre-evaluated accidents .

In one embodiment, the patient accident risk assessment method further includes the following steps: adjust each feature and pre-assessment according to the time point of the hospital visit, the time point of the accident or both on the medical record data The relevance of unexpected events.

In one embodiment, the patient accident risk assessment method further includes the following steps: using machine learning to use at least one of the multiple learning models to predict the algorithm, and analyzing the accident that has the highest correlation with the accident to be evaluated Characteristics to establish an optimal model.

In addition, a patient accident risk assessment system provided by the present invention includes a patient safety notification platform, a medical record database, an admission assessment database, and an accident risk assessment module. The patient safety notification platform stores the accidental data of each accident for each patient. The medical record database stores the medical record data of every patient who goes to the hospital for consultation. The admission assessment database stores admission assessment forms with multiple characteristics. The accident risk assessment module is connected to the patient safety notification platform, medical record database and admission assessment database. The accident risk assessment module performs the following procedures: Obtain multiple medical records with multiple characteristics and multiple accident data of pre-evaluated accidents; compare each feature with the multiple accident data based on multiple medical records and multiple accident data Pre-assess the relevance of unexpected events for feature screening; when performing feature screening, exclude features whose relevance is lower than the threshold of an excessively low relevance, and exclude features whose relevance is higher than the threshold of an excessively high relevance; use machines Learn to analyze the features that have the highest relevance to the unexpected event to be evaluated from the multiple features retained after feature screening to establish an optimal model; and based on the optimal model, based on the admission evaluation form filled out by newly admitted patients The multiple characteristics of the hospital assess the risk of accidents in newly admitted patients.

In one embodiment, in the patient accident risk assessment system, the accident risk assessment module is more implemented: using machine learning, from multiple features retained after feature screening, multiple features with relatively high relevance are analyzed The characteristics and the unexpected event to be evaluated are used to establish a training model; and machine learning is used to test the characteristics of the training model that have the highest correlation with the unexpected event to be evaluated to establish an optimal model.

In one embodiment, the accident risk assessment module of the patient accident risk assessment system retrieves multiple medical records within an assessment time range and multiple accident data of pre-evaluated accidents.

In one embodiment, the accident risk assessment module of the patient accident risk assessment system adjusts the correlation between each feature and the pre-evaluated accident based on the time point when each feature is generated, the time point when the accident occurs, or both. .

In one embodiment, the accident risk assessment module of the patient accident risk assessment system is further implemented: using machine learning to use at least one of multiple learning models to predict the algorithm, and analyze that the accident to be evaluated has the highest value. The characteristics of relevance to establish an optimal model.

In one embodiment, the accident risk assessment module of the patient accident risk assessment system regularly updates the optimal model.

As mentioned above, the patient accident risk assessment system and method of the present invention provide usable assessment items based on existing accident data and medical record data, and uses supervised machine learning to analyze these assessment items to establish an optimal model. It is applied to any medical institution to evaluate the risk of accidents for newly admitted patients, thereby reducing the probability of accidents for patients, thereby reducing the consumption of medical resources and improving the quality of medical services.

In order to further understand the features and technical content of the present invention, please refer to the following detailed description and drawings about the present invention. However, the provided drawings are only for reference and description, and are not used to limit the present invention.

The following are specific examples to illustrate the disclosed embodiments of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the concept of the present invention. In addition, the drawings of the present invention are merely schematic illustrations, and are not drawn according to actual dimensions, and are stated in advance. The following embodiments will further describe the related technical content of the present invention in detail, but the disclosed content is not intended to limit the protection scope of the present invention.

It should be understood that although terms such as “first”, “second”, and “third” may be used in this document to describe various elements or signals, these elements or signals should not be limited by these terms. These terms are mainly used to distinguish one element from another, or one signal from another signal. In addition, the term "or" used in this article should, depending on the actual situation, possibly include any one or a combination of more of the associated listed items.

Please refer to FIG. 1. FIG. 1 is a flowchart of steps of a patient accident risk assessment method according to an embodiment of the present invention; FIG. 2 is a chart of feature screening data of a patient accident risk assessment system according to an embodiment of the present invention. As shown in Figure 1, the patient accident risk assessment method of this embodiment includes steps S101 to S117. The whole process is divided into 5 parts, which are data acquisition and preparation, feature selection, use of machine learning, and sample training. And testing, the optimal model for dynamic risk assessment is used as a reference model for evaluating whether accidents occur in newly admitted patients, and the evaluation process of accidents in newly admitted patients is described in detail as follows.

First, in step S101, a patient safety reporting platform PSRS is used to obtain and store accidental event data for each accident that occurs for each patient. Each accidental event data can include the personal identification information of the patient who has the accident such as name, gender, age, etc., the accident such as a fall, the time of the accident such as year, month, day and time, and the location of the accident such as Area, address, impact of the accident on the patient, such as fractures, and treatment measures after the accident.

It should be understood that the patient safety notification platform PSRS is used to record accident data of accidents that have occurred, and the people who have these accidents do not necessarily go to the hospital for treatment. In other words, in practice, after an accident occurs, the person who has the accident may seek medical attention, but they may not. For the convenience of explanation, in this article, people who have accidents are collectively referred to as patients, but they should be interpreted as anyone who has accidents, for example, this person can be a so-called injury.

In step S102, the medical record database of each hospital is used to obtain and store the medical record (Electronic Health Record) data EHR of each patient in the same hospital. If multiple patients visit the same hospital, the medical record database stores the EHR of multiple patients' respective medical records. In this embodiment, the medical record data EHR is an electronic medical record, but the present invention is not limited to this.

For example, as shown in Figure 2, the medical record data EHR can include mobility, such as the patient can move independently, the patient needs to be supported by others, or the patient needs assistive tools such as crutches, wheelchairs, hospital beds or other transportation tools. . In addition, the medical record data EHR can include relevant characteristics such as whether the patient is being restrained by protection, using bed rails, etc. In addition, the medical record data EHR may include relevant characteristics such as whether to seek outpatient assistance, the type of clinic, treatment factors, the number of diagnoses, whether to take drugs for a specific purpose, and the type and amount of drugs taken. In addition, the medical record data EHR can include the patient’s age, gender, education, religious belief, marital status, and whether or not someone was accompanied at the time of consultation or admission. In addition, the medical record data EHR can include whether there have been accidents such as a fall within a time range (for example, one month), as well as the location of the accident, the patient's state of consciousness, the severity, the handling method, whether there is conflict, and the resulting concurrency Symptoms and other related features.

In step S103, each hospital can establish an admission assessment form AAS0 based on the health assessment items provided by it. The admission assessment form AAS0 may have multiple feature options for single selection of one feature or multiple selections, and may have blank fields for filling in one or more features. The admission assessment form AAS0 can optionally have accidents options for selecting accidents that have occurred or pre-assessment of accidents that may occur in the future. In this embodiment, the admission assessment form AAS0 shown in FIG. 1 is a blank form, that is, a form that is not provided for newly admitted patients to select or fill in.

In step S104, it is possible to retrieve historical accident data within the assessment time range, and retrieve the historical medical record data EHR of patients who have had accidents and visited the hospital within the assessment time range. For example, to retrieve accidental event data for N years, such as 9 years, for example, the evaluation time range is from 2011 to 2019. This is just an example, and the present invention is not limited to this.

It is worth noting that in the subsequent steps S105~S107, based on the multiple medical record data EHR and multiple accident data captured within the assessment time range, the characteristics and desires in the medical record data EHR are compared. The relevance of the assessed incidents.

Because of the low relevance to the pre-assessed unexpected event, the feature has no reference value. For example, among the N cases of falling events such as 500,000 historical medical records EHR, only 2 cases of patients have the characteristics of pale skin. If it is impossible to determine that the skin is pale, it will definitely fall, so pale skin cannot be regarded as The basis for judging whether a fall will occur. Therefore, in step S105, features whose relevance is lower than an excessively low relevance threshold (or lower than an excessively low threshold ratio) are excluded. As shown in Figure 1, p is defined as an estimated value, which is inversely proportional to the correlation, that is, the higher the correlation between the feature and the accident, the lower the p.

In this embodiment, features that meet the following formula are excluded: 1-p<0.9, where 1-p is defined as Feature importance in this embodiment. For example, as shown in Figure 2, characteristics such as gender, number of diagnoses, marital status, educational background, and religious beliefs are too low in association with unexpected events such as falls. As a result of the above formula, the generated feature importance is 0.774, 0.767, 0.345, 0.334, 0.086, each of which is lower than 0.9, so these features are excluded in step S105 and not used for subsequent model building.

Conversely, as shown in Figure 2, the patient’s mobility, the patient being restrained/using bed rails, seeking outpatient assistance, taking drugs that affect balance, the patient having a fall within a month, being older than 65, and being accompanied to hospital. The correlation with unexpected events such as a fall event is converted into an estimated value p that is inversely proportional to the correlation, and the resulting feature importance is 1, 1, 1, 1, 1, 1, 0.976 after the above formula is used. Each is higher than 0.9, so these features are temporarily retained in step S105.

Furthermore, due to the feature of excessively high correlation with the pre-assessed accident, even if there is this feature, an accident will definitely occur. For example, as long as the knee has flushing, it will definitely (100%) fall. This theory is not reasonable. It has no reference value. Therefore, in step S106, the features whose relevance is higher than the too high association threshold (or the too high threshold ratio) are excluded, that is, the overfit features are excluded. As shown in Figure 1, r is the correlation, that is, the higher the correlation between the feature and the accident, the higher the r.

In this embodiment, features that meet the following formula are excluded: r>0.95, where 0.95 is an excessively high correlation threshold, that is, the proportion of excessively high threshold is 95%. For example, as shown in FIG. 2, the accidental event retained after the execution of step S105, for example, the amount of data N falling within a time range (for example, within 1 month) is 391, and all the data in this time range The total amount of accidental data N is 405, and the calculated correlation is 96.5%, which is higher than 0.95 of the above formula. According to this, as long as there is a fall, almost (96.5%) will fall again within 1 month. This theory does not It is reasonable and has no reference value, so it excludes the characteristics of falling history. In addition, if a patient has been diagnosed in the hospital due to an accident such as a fall, the chance of another accident such as a fall is 99.5%, which is higher than 0.95 in the above formula. According to this, as long as the patient has been diagnosed in the hospital due to a fall, he will basically Once again, this theory is unreasonable and has no reference value.

More precisely, the correlation between the feature and the pre-assessed accident depends on the probability of the accident that has the feature of the patient, and the feature can also depend on the time point, location and cause of the accident. For example, the relevance between the accident data from 2011 to 2019 and the medical record data EHR captured in step S104 is multiplied by different weight values according to different years, and this embodiment is multiplied by different weight values every two years, for example The relevance from 2018 to 2019 is multiplied by 100% (that is, the weight value is not multiplied), the relevance from 2016 to 2017 is multiplied by 80%, the relevance from 2014 to 2015 is multiplied by 60%, and so on This is only an example, and the present invention is not limited thereto.

After performing feature screening through steps S105 and S106 as shown in Figure 1, in step S107, the remaining features that meet the feature selection are obtained, such as the patient’s mobility, constrained/used bed rails, The characteristics of outpatient assistance, medication taking, age, and accompanying admission are used as the analysis characteristics of machine learning in the following subsequent steps.

In step S108, the feature values retained after the above feature screening are used as the machine learning samples for model building. The data is randomly allocated to two mutually exclusive data sets, 75% of the data is used to build the model, and the remaining 25% is used for the model Test and select the best model according to the method of performance measurement. For example, as shown in Figure 2, there are a total of 405 data records, of which 75% (290 records) of the data are used to build the training model, and the remaining 25% (115 records) are used for model testing.

In the 75% training (Training) grouping in step S109, machine learning is used to analyze the multiple symptoms that have relatively high correlation with the unexpected event to be evaluated from the multiple symptoms retained after feature screening, so as to Step S110 establishes a training model.

In the 25% testing (Testing) clustering in step S111, machine learning is used, such as LR, BN, ANN or CHAID algorithms, to test the training model that has the highest correlation with the unexpected event to be evaluated. In step S112, an optimal model is established for evaluating the accident risk of newly admitted patients in subsequent steps S113 to S117.

In step S103 shown in Fig. 1, the hospital establishes an admission assessment form AAS0. In step S113, the hospital provides the admission assessment form AAS0 for the newly admitted patient to fill in and check. For example, a patient selects his current symptoms from a large number of disease options, and a newly admitted patient fills in the above-mentioned characteristics such as name, age, marital status, etc. into the corresponding fields on the admission assessment form AAS0. To obtain the completed admission assessment form AAS1.

In step S114, based on the optimal model, an accident prediction model is generated based on one or more features filled in and checked in the admission assessment form AAS1 of the newly admitted patient. For example, the patient accident risk assessment method can provide multiple learning model prediction algorithms, such as the Fall Prediction Model (Falls Down Prediction Mode) FPM, the Morse Fall Scale (MFS), and the Hendrich Falls Assessment Tool (Hendrich Falls). Assessment Tool, HFAT), Thomas’ Risk Assessment Tool in Falling Elderly Inpatients (STRATIFY), etc. Each hospital can use machine learning to use multiple learning models to predict at least one of the algorithms according to actual application requirements, and analyze the features that have the highest correlation with the unexpected event to be evaluated to establish an optimal model.

In step S115, the accident prediction model predicts whether an accident occurs, as a basis for medical staff to evaluate whether an accident has occurred in the patient. For example, the accidental event evaluated in the optimal model is a fall, which can be used as the evaluation of the characteristic value that is most likely to cause a fall to be a fracture. When the characteristic value of the fracture is checked or filled in on the admission assessment form AAS1, the accident prediction model points out that the patient will fall, and/or informs the patient of the probability of falling. When multiple different characteristic values are checked or filled in on the admission assessment form AAS1, a comprehensive assessment of possible accidents should be made.

In step S116, if the accident prediction model predicts that an accident will occur with a high probability, corresponding safety care measures can be drawn up based on the accident evaluated in the accident prediction model. For example, the accidental event evaluated by the accident prediction model is a fall. Nursing staff can put a label with fall information on the patient or the vicinity of the patient's bed, or use a protective device to fix the fractured leg of the patient.

In step S117, the time range of the accident assessment can be further set, such as assessing whether an accident will occur in the first month, the second month, or the third month and the probability of occurrence.

Please refer to FIG. 3, which is a block diagram of a patient accident risk assessment system according to an embodiment of the present invention. As shown in FIG. 3, the patient accident risk assessment system of this embodiment includes a patient safety notification platform 10, a medical record database 20, an admission assessment database 30, and an accident risk assessment module 40, which uses the aforementioned patient accident risk assessment method.

The accident risk assessment module 40 is connected to the patient safety notification platform 10, the medical record database 20, and the hospital admission assessment database 30. The accident risk assessment module 40 is configured to evaluate the accident risk of the patient based on the accident data provided by the patient safety notification platform 10, the medical record data provided by the medical record database 20, and the admission assessment form provided by the hospital admission assessment database 30 Sex.

Please refer to FIGS. 3 and 4 together. FIG. 4 is a detailed block diagram of an optimized model of the patient accident risk assessment system according to an embodiment of the present invention.

The hospital HS described in this article covers various medical institutions. The hospital HS provides n patients PAT1~PATn to the hospital HS for treatment, and the medical record database 20 of the hospital HS stores multiple medical records EHR1~EHRn of the n patients PAT1~PATn who have visited the hospital HS. The medical records EHR1~EHRn contain the characteristics of each patient's PAT1~PATn. Among them, n is any positive integer.

Hospital HS can provide admission assessment form AAS0 to patients PAT1~PATn. The admission assessment form AAS0 can have various feature options or fill in fields for each patient PAT1~PATn to fill in or check j features FET1~FETj, where j is the ranking number of each feature, which can be any positive value greater than 1. Integer. After n patients PAT1~PATn have filled out an admission assessment form AAS0 respectively, they will submit the admission assessment form AAS0 to the hospital HS.

The medical record data EHR1~EHRn of the hospital HS can include the patient’s PAT1~PATn filled or checked on the admission assessment form AAS0, and one or more characteristics FET1~FETj diagnosed by the hospital HS. For example, the medical record data EHR1 of the patient PAT1 includes multiple features FET1, FETj such as headaches and fractures, while the medical record data EHRn of the patient PATn includes multiple features FET7, FET9 such as dizziness and walking bumps.

The patient PATi~PATm may have any one or more of the multiple characteristics FET1~FETj, leading to accidents AC1~ACm. After the accident AC1~ACm of the patient PATi~PATm, the patient safety notification platform 10 stores (m-i+1) accident data ACID1~ACIDm of the patient PATi~PATm respectively, where i is any positive integer value; m represents The number of patients with accidents, which is a positive integer value greater than i.

It should be understood that even if each patient PAT1~PATn attending the hospital HS has any one or more of the characteristics FET1~FETj, it does not mean that any one or more accidents AC1~ACm must have occurred. Generally speaking, the number n of patients PAT1~PATn visiting the hospital HS is more than the number of patients PATi~PATm visiting the hospital HS due to accidents.

For example, if an accident ACm occurs in a patient's PATm but does not go to the hospital HS for treatment, the patient safety notification platform 10 will record the accident data ACIDm of the patient's PATm, but the patient's PATm will not occur in the hospital's HS medical record database 20 This accident ACm led to the medical record data of the injury.

Conversely, if the patient PATn has no accident but goes to the hospital HS for treatment, the medical record database 20 of the hospital HS stores the medical record data EHRn of the patient's PATn, but the patient safety notification platform 10 will not have the accidental event data of the patient's PATn.

If an accident ACm6 occurs in the patient PAT6 and there is a visit to the hospital HS, the patient safety notification platform 10 and the medical record database 20 of the hospital HS will store the accident data ACID6 and the medical record data EHR6 of the patient PAT6, respectively.

Furthermore, the accident risk assessment module 40 can assess the relevance of all the features FET1~FETj on the admission assessment form AAS0 filled in within a time frame to the pre-assessed accident ACx, where x is a positive integer less than m . It is worth noting that, as in the above-mentioned steps S105 to S107, feature screening is performed to exclude the features that are too high or too low in correlation with the pre-evaluated accident ACx among the multiple features FET1 to FETj.

The accident risk assessment module 40 uses machine learning to analyze the multiple features FET2, FET3, FET8, FET11, FET15 that have the highest correlation RV6 with the accident ACx from the multiple symptoms FET1~FETj or less retained after feature screening. For example, the correlation between these features RV6 and the accident ACx is 92% to establish the optimal model MSM. The optimal model MSM can be updated regularly, for example, the optimal model MSM is updated every few years, for example, every two years, that is, the optimal model MSM is re-established based on a different evaluation time range. This is only an example for illustration, and the present invention is not limited thereto.

Please also refer to FIG. 5, which is a detailed block diagram of an accident prediction model of a newly admitted patient established based on an optimal model of the patient accident risk assessment system according to an embodiment of the present invention.

It should be understood that the patients PAT1~PATn shown in Figure 4 refer to the old patients who have been to the hospital HS for consultation, and their time to see the hospital HS is earlier than that of the newly admitted patient NEWPAT as shown in Figure 5 The time of the HS visit. The accident risk assessment module 40 establishes an optimal model based on the medical record data EHR1~EHRn of the patients PAT1~PATn who have been to the hospital HS, and the accident data ACID1~ACIDm of the patients PATi~PATm who had accidents AC1~ACm MSM. As shown in Figure 5, the accident risk assessment module 40 stores the optimized model MSM established above, which is used to establish the basis for establishing the accident prediction model ACPM of NEWPAT for newly admitted patients as follows to evaluate the accidental events encountered by NEWPAT of newly admitted patients Risks, improve the quality of medical services, as detailed below.

First, after the newly admitted patient NEWPAT enters the hospital HS counter to register or make an online appointment to see the hospital HS, the hospital HS provides a blank admission assessment form AAS0 through the admission assessment database 30 to the newly admitted patient NEWPAT, this admission assessment form AAS0 can be an electronic form or paper. There are multiple feature options on the admission assessment form AAS0 for tick, circle or fill in one or more features FET1~FETj, where j represents the sequence number of the feature, which is any positive integer greater than 1.

For example, a newly admitted patient NEWPAT selects and fills in the characteristics FET5, FET6, and FET11 on the admission assessment form AAS0. After the newly admitted patient NEWPAT fills out the admission assessment form AAS0, the admission assessment form AAS1 will be generated and provided to the accident risk assessment module 40. The accident risk assessment module 40 is based on the optimal model MSM, and generates an accident prediction model ACPM based on the characteristics FET5, FET6, and FET11 selected and filled in by the newly admitted patient NEWPAT.

The accident prediction model ACPM can evaluate the accidents that are likely to occur within a time range in the future based on these characteristics FET5, FET6, and FET11 of the NEWPAT of newly admitted patients, and even inform the probability of accidents.

Further, the nursing staff of the hospital HS can view the accident prediction model ACPM of the NEWPAT of the newly admitted patient under the care of the accident risk assessment module 40 of the patient accident risk assessment system of the hospital HS, and provide safe care measures SCM accordingly. In practice, the accident risk assessment module 40 can further indicate possible accidents based on the accident prediction model ACP, and instruct the nursing staff to provide corresponding safety care measures SCM to the newly admitted patients NEWPAT to prevent actual accidents from occurring. After the newly admitted patient NEWPAT visits the hospital HS, a new electronic medical record NEWEHR will be generated and stored in the medical record database 20. The new electronic medical record NEWEHR can be used as the basis for establishing the optimal model for future updates.

Please refer to FIG. 6, which is a flowchart of the steps of the security notification method according to an embodiment of the present invention.

First of all, it should be understood that the transmission of information about safety hazards suffered by newly admitted patients in the hospital is mainly based on the safety notification mechanism of newly admitted patients. In Taiwan's newly admitted patient safety notification incidents, nurses are the main informants. As far as the main users of the safety notification method are concerned, nurses are a key role in promoting the safety and quality of newly admitted patients. The study pointed out that medical staff are not clear about the importance of the safety notification of newly admitted patients, but worry about being punished because of the notification. As a result, the number of hospital notifications in Taiwan is low. By changing the hospital management culture and reporting the issue of impunity, and establishing effective and safe reporting information methods, the purpose of medical quality services can be improved.

Therefore, this embodiment establishes a user-centered method for the safety notification information of newly admitted patients in the hospital, which can be applied to the patient safety notification information system of various medical institutions, providing nursing staff can easily achieve the notification purpose in a specific use environment. Effective safety notification of newly admitted patients is very important for improving the safety of newly admitted patients.

As shown in FIG. 6, the method for safety notification of newly admitted patients in the hospital of this embodiment includes steps S601 to S622, which are divided into quick notification and complete notification. Among them, performing steps S601 to S614 can realize quick notification. In the rapid notification process, you only need to log in to the system, select the unit’s newly admitted patients and report the incident, and check the location, cause, severity, and consciousness of the newly admitted patient after the incident, generate a notification serial number, and proceed Notice in the hospital. After the person who is to be notified is finished caring for the patient, perform steps S615~S622 to fill in the detailed content of the post-event report to realize a complete notification. In this way, the informant can inform the supervisor of the incident information in a short time without affecting the urgency of the informant to take care of the newly admitted patients.

That is, first, the quick notification process of steps S601 to S614 is performed.

In step S601, when a newly admitted patient is admitted to the hospital after an accident, the informant, such as a nursing staff responsible for caring for the newly admitted patient, can use the patient safety notification information system.

In step S602, the hospital can be divided into multiple medical units, such as internal medicine, surgery, obstetrics and gynecology, pediatrics, oncology, neurological rehabilitation, psychiatry, emergency and intensive surgery and supply rooms. Nursing staff can enter their name and password on the system interface to log in to the main page of rapid notification of medical units that can diagnose and handle symptoms caused by accidents in newly admitted patients.

In step S603, after logging in the main page of the quick notification of the medical unit of the safety notification information system, the nursing staff can select the data of the newly admitted patient from a number of options, such as the name of the newly admitted patient and the accident items that occurred. Enter the sub-page.

In step S604, on the quick notification sub-page of the medical unit, the nursing staff can select the location of the accident, the cause, the severity of the symptoms, and the awareness of the newly admitted patient from multiple options. status.

In step S605, the nursing staff can select one or more notification methods from multiple notification methods, such as Short Message Service (SMS), E-mail, etc.

In step S606, the patient safety notification information system integrates the personal identification data of the newly admitted patient and the accident related data, and in step S607, stores the information of the newly admitted patient in the event database.

Nursing staff can obtain the personal identification data of newly admitted patients and accidental event-related data, and record these data in the event database, that is, after completing steps S601~S607 and other required procedures for registration, they can inform the newly admitted patient to complete the registration , And then in step S608, the registration data of newly admitted patients can be stored in the registration database.

In step S609, the safety notification information system judges whether the accident of the newly admitted patient is a new event. If the accident of the newly admitted patient is not a new event, step S611 is executed to search for the serial number of the accident of the newly admitted patient from the event database, that is, use the existing accident serial number. Conversely, if the unexpected event of the newly admitted patient is a new event, step S610 is executed to generate a new sequence number of the unexpected event, which can be sequenced after the sequence number of the old event.

In step S612, the notification database of the safety notification information system can store the personal identification data of the newly admitted patient and the accident (the serial number) of the incident.

In step S613, the safety notification information system can adopt the notification method selected in step S605 to notify the newly admitted patient's personal identification data stored in the database and the accident (the serial number) of the patient, etc., and actively notify the responsible caregiver The nursing staff of the newly admitted patient (that is, the informant who performs the above steps), the direct chief of this nursing staff, and other persons who should be notified, such as the nursing chief, director, etc., can complete the quick report.

Furthermore, after taking care of the newly admitted patient, the nursing staff performs the following steps S615 to S622 to complete the notification.

The person who obtained the notification in step S614, such as the nursing staff caring for this newly admitted patient, can operate the security notification information system to enter the name, password and select the medical unit in step S616 to log in to the medical unit that the newly admitted patient sees. Full notification page.

In step S617, the nursing staff can select the serial number of the accident occurred in the newly admitted patient on the interface of the safety notification information system to view/check the information of the newly admitted patient. If necessary, the data content can be added or modified in step S618.

In step S619, the nursing staff can reply to the newly admitted patient's comments on the service quality of the hospital.

In step S620, the nursing staff can choose whether to exit the system page. When exiting, the information and comments of the newly admitted patients in the above steps can be saved in the event database.

In step S621, the nursing staff can determine whether to complete safety care measures to prevent accidents from recurring accidents of newly admitted patients, and to appropriately deal with symptoms caused by accidental measures, and choose whether to end the modification of the patient's data content. If so, the safety notification information system can store patient data for viewing by authorized personnel in the hospital. Compared with quick notification, more people are notified, so that a complete notification can be realized in step S622.

[Beneficial effects of the embodiment]

The beneficial effect of the present invention is that the patient accident risk assessment system and method of the present invention provide usable assessment items based on existing accident data and medical record data, and use supervised machine learning to analyze these assessment items to establish an optimal The model can be applied to any medical institution to assess the risk of accidents for newly admitted patients, thereby reducing the probability of accidents for patients, thereby reducing the consumption of medical resources and improving the quality of medical services.

The content disclosed above is only a preferred and feasible embodiment of the present invention, and does not limit the scope of the patent application of the present invention. Therefore, all equivalent technical changes made using the description and schematic content of the present invention are included in the application of the present invention. Within the scope of the patent.

S101~S117, S601~S622: steps N: data volume X: Parameters p: estimated value 10. PSRS: Patient Safety Notification Platform 20: Medical records database 30: Admission assessment database 40: Accidental Risk Assessment Module HS: Hospital AAS0, AAS1: admission assessment form FET1~FETj, FET2, FET3, FET5, FET6, FET7, FET8, FET9, FET11, FET15: Features PAT1~PATn, PATi~PATm: patient EHR, EHR1~EHRn: medical record data AC1~ACm: accident ACID1~ACIDm: accident data RV1~RVj, RV6: correlation MSM: Optimal Model ACx: accident SCM: safety care measures NEWPAT: Newly admitted patients NEWEHR: New electronic medical record ACPM: unexpected prediction model FPM: Fall prediction model

Fig. 1 is a flow chart of the steps of a method for assessing patient accident risk according to an embodiment of the present invention.

Fig. 2 is a chart of feature screening data of the patient accident risk assessment system according to an embodiment of the present invention.

Fig. 3 is a block diagram of a patient accident risk assessment system according to an embodiment of the present invention.

Fig. 4 is a detailed block diagram of an optimized model of the patient accident risk assessment system according to an embodiment of the present invention.

Fig. 5 is a detailed block diagram of an accident prediction model of a newly admitted patient established based on an optimal model of the patient accident risk assessment system according to an embodiment of the present invention.

FIG. 6 is a flowchart of the steps of a method for security notification of information according to an embodiment of the present invention.

S101~S117: steps

Claims (12)

A patient accident risk assessment method, including the following steps: use a patient safety notification platform to store an accident data of each accident for each patient; use a medical record database to store a medical record of each patient in the hospital Establish an admission assessment form, the admission assessment form has multiple characteristics; obtain the multiple medical records with the multiple characteristics and the multiple accident data of the pre-evaluated accidents; based on the multiple medical records and the The data of multiple accidents are compared, and the correlation between each feature and the accident to be evaluated is compared for feature screening; the multiple correlations between the multiple features and the accident to be evaluated are based on the generation of the multiple Different years of features are multiplied by different weight values; the correlation between each feature and the unexpected event to be evaluated is converted into an estimated value that is inversely proportional to the correlation; when performing feature screening, exclude those that meet the following formula Feature: 1-p<0.9, where p is the estimated value to exclude the feature with too low relevance; when performing feature screening, exclude the feature that meets the following formula: r>0.95, where r is the relevance, and Exclude the feature with high relevance; use machine learning to analyze the feature with the highest relevance to the unexpected event to be evaluated from the multiple features retained after feature screening, so as to establish an optimal model; On the admission evaluation form, based on the symptoms of the newly admitted patient, check from the multiple characteristics of the admission assessment form; and based on the optimal model, according to the newly admitted patient checked on the admission assessment form. Or multiple characteristics to assess the risk of accidents in newly admitted patients. For the patient accident risk assessment method described in item 1 of the scope of patent application, the steps performed by machine learning include: Using machine learning, from the multiple features retained after feature screening, the multiple features that have relatively high relevance to the unexpected event to be evaluated are analyzed to establish a training model; and using machine learning to test out In the training model, the feature that has the highest correlation with the unexpected event to be evaluated is used to establish the optimal model. For example, the patient accident risk assessment method described in item 1 of the scope of patent application further includes the following steps: setting an assessment time range; and extracting the multiple medical records within the assessment time range and the number of pre-assessed accidents Accident data. For example, the patient accident risk assessment method described in item 1 of the scope of the patent application further includes the following steps: adjust each feature and the time point of the accident according to the time point of the hospital visit, the time point of the accident or both on the medical record data The relevance of pre-evaluated unexpected events. For example, the patient accident risk assessment method described in item 1 of the scope of the patent application further includes the following steps: using machine learning to predict at least one of the algorithms using multiple learning models, and analyzing the highest relevance to the accident to be evaluated The characteristics of the, to establish the optimal model. The patient accident risk assessment method described in item 1 of the scope of patent application further includes the following steps: regularly updating the optimization model. A patient accident risk assessment system, including: a patient safety notification platform, which stores one accident data for each accident for each patient; a medical record database, which stores one medical record data for each patient who is sent to the hospital; Admission assessment database, which stores an admission assessment form with multiple characteristics; An accident risk assessment module, which connects the patient safety notification platform, the medical record database, and the hospital admission assessment database. The accident risk assessment module performs the following procedures: obtaining the multiple medical records with the multiple characteristics and pre-assessment According to the multiple accident data of the accidents; according to the multiple medical records and the multiple accident data, compare the correlation between each of the characteristics and the pre-assessed accidents for feature screening; The multiple correlations between a feature and the unexpected event to be evaluated are multiplied by different weights according to the different years in which the multiple features are generated; the correlation between each feature and the unexpected event to be evaluated is converted into a correlation An estimated value that is inversely proportional; when performing feature screening, exclude the feature that meets the following formula: 1-p<0.9, where p is the estimated value to exclude the feature with too low relevance; when performing feature screening, Exclude the feature that meets the following formula: r>0.95, where r is the relevance to exclude the feature that is too highly relevance; using machine learning, from the multiple features retained after feature screening, analyze the features that are relevant to the evaluation The feature with the highest relevance of accidents is used to establish an optimal model; on the admission assessment form, based on the symptoms of the newly admitted patient, check from the multiple characteristics of the admission assessment form; and based on the optimal According to the one or more characteristics of the newly admitted patients selected on the admission assessment form, the risk of accidents of the newly admitted patients is evaluated. For example, the patient accident risk assessment system described in item 7 of the scope of patent application, wherein the accident risk assessment module is more implemented: using machine learning, from the multiple features retained after feature screening, analyze the unexpected events to be evaluated The multiple features with relatively high relevance to establish a training model; and Using machine learning, the feature that has the highest correlation with the unexpected event to be assessed in the training model is tested to establish the optimal model. For example, in the patient accident risk assessment system described in item 7 of the scope of patent application, the accident risk assessment module retrieves the multiple medical record data and the multiple accident data of the pre-evaluated accident within an assessment time range. For example, the patient accident risk assessment system described in item 7 of the scope of patent application, wherein the accident risk assessment module adjusts each feature and the pre-assessment based on the time point when each feature is generated, the time point when the accident occurs, or both The relevance of unexpected events. For example, in the patient accident risk assessment system described in item 7 of the scope of patent application, the accident risk assessment module is further implemented: using machine learning to use at least one of multiple learning model prediction algorithms to analyze the accident to be evaluated The feature with the highest relevance of the event is used to establish the optimal model. For example, in the patient accident risk assessment system described in item 7 of the scope of patent application, the accident risk assessment module regularly updates the optimization model.

Images