JP2006031445A - Accident, incident report analysis system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To exactly grasp relationships between an accident, an incident, a nurse or environment by analyzing through complemented or interpolated data mining. <P>SOLUTION: When there is lacked information in free description of a report, a nursing activity monitor system is referred to and the lacked information is complemented or interpolated (drawing (A)) by combining an integrated nursing activity monitor system 10, a nursing accident and an incident report. Meanwhile, when information on the accident or the incident is missing in the nursing activity monitor system or not followed well, the accident and incident report is referred to and the lacked information is complemented or interpolated (drawing (B)). Data complemented or interpolated is analyzed by a data mining technique. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  TECHNICAL FIELD The present invention relates to an accident / incident report analysis system, and in particular, a medical (nursing) accident (accident) or incident (incident) report by a doctor or a nurse (hereinafter referred to as an “accident / incident report”). .) Analyzing accident and incident report analysis system.

  Many hospitals have been accused of medical (nursing) mistakes and have to pay a large amount of compensation after losing them. Therefore, some hospitals insure medical (nursing) mistakes to prevent bankruptcy. Of course, medical (nursing) mistakes are not just a problem for hospitals. It also has a serious impact on patients. Insurance can prevent bankruptcy. However, it cannot save the patient's life. In other words, it is very passive in terms of preventing medical (nursing) mistakes.

  Recently, it has been recognized that medical risk management is very important for both hospitals and patients. Medical risk management seeks to minimize nursing costs and insurance. Therefore, it seems better to lower the insurance money. To that end, it is important to reduce nursing accidents and incidents. By doing so, it will be for the hospital and for the patient.

  In order to avoid nursing accidents and incidents, it is important to analyze cases of nursing accidents and incidents that occurred previously. If you collect and generalize what happened before, keep in mind that you can avoid the same or similar. Therefore, hospitals usually make accident and incident reports to investigate why nursing accidents happened.

  Reports are usually read by those who specialize in risk management, analyze accidents and incidents, and create cases and textbooks on nursing risk management. Unfortunately, it takes a lot of time to read all the reports and extract and generalize important points.

  From the viewpoint of data mining (data mining), considering that some hospitals have recently introduced electronic medical records, it is best to analyze a large number of reports with the help of computers. Looks like.

  Data mining technology has been used to analyze sales trends and economic trends and to predict future situations from the current situation. In this way, data mining technology is used to find current trends and past and future trends. Therefore, if the data mining technique is applied to analysis of nursing accidents and incidents, it is possible to search for trends in nursing accidents and incidents, and to estimate typical nursing accidents and incident patterns. With typical patterns, nurses can avoid many nursing incidents and incidents that are likely to occur in the future.

  Part of the data mining method (decision tree generation) has already been used for analysis of nursing accidents and incident reports, as shown in Non-Patent Document 1.

  In this non-patent document 1, data mining from an accident / incident report (report obtained from an electronic medical record system) was performed. Specifically, in this non-patent document 1, ICONS-Miner (trade name) (infectious disease test data mining support software, manufactured by Koden Kogyo Co., Ltd.) (see: http://www.koden.jp/product_02. html) was used to generate decision trees. Some results obtained in Non-Patent Document 1 are shown in Table 1.

Park Kusunoki, Yurai Hirai, Saeko Ono, Toshimitsu Fujii, Yoshiki Nishizawa: Analysis and Practice of Safety Management Report by Data Mining Method, Proc. Of 22nd JCM, pp.712-713, 2002

  In the analysis in Non-Patent Document 1, as shown in Table 1, trends and rules have been discovered. However, they are very typical trends and rules that have already been pointed out by Rix Management Experts. Furthermore, these trends are very superficial. The reason is just an experiment to find out how much data mining can be used to analyze accident and incident reports, and only the options in the reports were analyzed. Since the choices and the fixed input part contain only predictable information, the result of Non-Patent Document 1 is rather simple and seems like a natural result.

  Therefore, a main object of the present invention is to provide a novel medical accident information analysis system.

  Another object of the present invention is to provide an accident / incident report analysis system capable of analyzing a detailed and accurate relationship between an accident / incident and a nurse / environment.

  The invention of claim 1 is an accident / incident report analysis system for analyzing a nursing accident / incident report by data mining, and when analyzing the free description of the report, the missing information is referred to the nursing activity monitoring system. It is an accident and incident report analysis system that complements it.

  The invention according to claim 2 is the accident / incident report analysis system according to claim 1, wherein information missing in the nursing activity monitoring system is supplemented by referring to the report.

  The invention according to claim 3 is the accident / incident report analysis system according to claim 1 or 2, wherein the information of the nursing activity monitoring system is analyzed by data mining.

  According to the present invention, the missing or missing information is supplemented by referring to the nursing activity monitoring system, and the free description of the accident and incident report is analyzed by data mining. Analyzes detailed and accurate relationships between teachers and environments.

  The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

  Referring to FIG. 1, an integrated nursing activity monitoring system 10 according to an embodiment of the present invention is applied to, for example, a hospital and includes a server 12. The server 12 is connected to a plurality of nurse terminals (hereinafter simply referred to as “terminals”) 16 and a plurality of stations 18 via a wired or wireless communication line (network) 14. The terminal 16 is a computer such as a personal computer or a workstation, and is assigned to each nurse. For example, the terminal 16 is installed in a nurse's station. However, a single terminal 16 may be used by several nurses. Each of the plurality of stations 18 is in a ward that accommodates inpatients, and is disposed in predetermined positions such as a corridor, a hospital room (entrance, bed, or the vicinity thereof), and a nurse's station, and is capable of wireless communication. Hereinafter, the identification information of 20 is simply acquired and transmitted to the server 12. Each of the plurality of sensor units 20 is assigned (attached) to a nurse, and the identification information of the sensor unit 20 is detected by the station 18 that exists in a wirelessly communicable range (for example, a radius of 3 to 5 meters). . Further, since the station 18 and the sensor unit 20 can communicate with each other, the sensor unit 20 detects identification information (station ID) of the station 18 existing in a range where wireless communication is possible. Although illustration is omitted, since the sensor units 20 can wirelessly communicate with each other, the two or more sensor units 20 existing in a wireless communicable range detect mutual identification information. You can also. The sensor unit 20 may be directly connected to the network 14 by a wireless LAN.

  FIG. 2 is a block diagram showing a specific configuration of the server 12. Referring to FIG. 2, the server 12 includes a plurality of objective sensors 22, a plurality of sound collecting microphones 24, which are omitted in FIG. A plurality of load sensors 25 are connected. Although not shown, in this embodiment, the plurality of objective sensors 22 stores a medical instrument such as a thermometer, a sphygmomanometer, a syringe, a drip syringe (injector), a blood collection test tube, and a urinalysis cup. It is installed in a box take-out part or a nurse call terminal and detects the use (take-out) of those medical devices. However, identification information such as tag information or a barcode may be added to these medical instruments and the presence or absence of use may be detected by reading the tag information or barcode. The plurality of sound collecting microphones 24 are installed in a ward that accommodates inpatients, and are installed in a corridor, a hospital room (a bed or the vicinity thereof), and collect ambient sounds. Further, the plurality of load sensors 25 are installed on the handrail of the bed used by the inpatient.

  In addition, when using video data for data mining and data complement described later, it is necessary to install a video camera (moving image and / or still image) in order to acquire such video data. For example, it will be provided in a nursery, a corridor, a hospital room, etc.

  Although omitted in FIG. 1, a plurality of databases (DB) are connected to the server 12. Specifically, a nurse DB 26, a ward event DB 28, a location information DB 30, an electronic medical record DB 30, and a knowledge DB 30 are connected.

  The nurse DB 26 records data (nurse data) on the behavior of the nurse in one day of work for each nurse. In the nurse data, for example, a nurse name is described as a label, and a date is described next to the nurse name. Below the nurse name and date, items (contents) related to the nursing work of the nurse are described. In other words, the start time of nursing work, the end time of nursing work, the target patient of nursing work, the contents of nursing work, the number of steps in nursing work, the average value of the inclination (inclining the upper body (body)) and the medical error in nursing work Accidents (accidents / incidents) in business such as Such nurse data is basically obtained from data measured by the sensor unit 20.

The ward event DB 28 includes a plurality of sensors, that is, object detection data obtained from the plurality of object sensors 22 described above, sound data obtained from the plurality of sound collecting microphones 24, and load detection data obtained from the load sensor 25 (collecting these data). Record the “ward event data”). Identification information such as the name of a medical instrument to be detected is attached to the object detection data as a label. Identification information indicating the installation location of the sound collecting microphone 24 is attached to the sound data as a label. Identification information indicating the installation location of the load sensor 25 is attached to the load detection data as a label. That is, the detection target can be specified from the object detection data, the location where the sound is collected can be specified from the label of the sound data, and the bed can be specified from the load detection data. Further, the server 12 acquires the time (time data) at which the ward event data is acquired from the built-in clock circuit 12a, and adds the acquired time to the ward event data.

  For example, when objective detection data is input, the server 12 adds time data to the objective detection data and records it in the ward event DB 28. Since the sound collecting microphone 24 is mainly intended to collect the falling sound of an object, when sound of a predetermined level or higher is input, sound data corresponding to the sound is input to the server 12, The server 12 adds time data to the sound data and records it in the ward event DB 28. When load detection data is input, the server 12 adds time data to the load detection data and records it in the ward event DB 28.

  In the location information DB 30, location information data (location information data) about the nurse is recorded. As will be described in detail later, the location information data includes identification information (nurse ID) such as tag information assigned to the sensor unit 20 worn by each nurse and the station 18 that detects the nurse ID. The tag information, that is, the identification information (station ID) is recorded in time series. Further, as described above, since the station 18 and the sensor unit 20 communicate with each other, the location information data includes data in which the station ID detected by the sensor unit 20 is recorded in time series. Further, as described above, the sensor units 20 can communicate with each other wirelessly, and the nurse ID assigned to each sensor unit 20 and the nurse ID of the sensor unit wirelessly communicated with the sensor unit 20 are recorded in time series. Data is also included. With this location information data, the location (action) of the nurse can be known. For example, you can know where you were in what hour, minutes, seconds (patient room, station, corridor, etc.), and what nurses you were with or passed in what hours, minutes, seconds it can.

  In the electronic medical record DB 32, an electronic medical record about a patient currently hospitalized or being treated is recorded. Although illustration is omitted, this electronic medical chart is obtained by extracting necessary items from the contents of the medical chart described by a doctor or the like and adding necessary items to electronic data. For example, the electronic medical record contains data (text data) such as patient name, nurse name in charge (nurse category number), gender, disease name, medical condition, prescription, rehabilitation status, hospitalization calendar, and patient precautions. In addition to being described, the patient category number (k) for the patient is attached to the text data as a label. Therefore, when a patient is specified, an electronic medical record for the patient can be specified according to the patient category number (k).

  In the knowledge DB 34, electronic medical records about patients who have been hospitalized or treated in the past are recorded, and medical accidents that occurred or were likely to occur in the nursing of such patients or actions of such patients Examples of accidents that occurred based on this are stored. Also, based on the results of examining such cases, the causes of accidents (including accidents that were likely to occur) are pursued (explored), and precautions for nurses and patients to avoid accidents in advance are provided. The contents are stored.

  For example, a fall accident occurred when a patient with lower limb paralysis due to head injury started standing on his own and tried to pick up an object dropped under his bed. This suggests that patients with the symptom (lower leg paralysis due to head injury) should be urged to call a nurse with a nurse call if there is a fallen object. Can be acquired. If there is a fallen object, the presence or absence of a nurse call is detected. If there is no nurse call, the nurse calls the patient or visits the patient. It is also possible to acquire knowledge that accidents can be avoided. Such cases and knowledge are built in the knowledge DB 34.

  In the knowledge DB 34, text data is written for a manual that describes the instruments used for nursing work and the correct progress of the nursing work. For example, if the nursing service is “thermometer”, the instrument used is a thermometer, (1) visit the patient, (2) hand the thermometer to the patient, and (3) record the result of the thermometer. It is recorded. In addition, if the operation of the medical instrument (device) is troublesome, a correct operation method for the device is also included.

  FIG. 3 is a block diagram showing a specific configuration of the sensor unit 20. With reference to FIG. 3, the sensor unit 20 includes a CPU 40. A / D converters 42 and 44, an encoder 46, a non-contact sensor 48, an interface 50, a card slot 52, a clock circuit 54, a DIP switch 56, a wireless transmitter 58, and a wireless receiver 60 are connected to the CPU 40. A pedometer 62 is connected to the A / D converter 42, and the A / D converter 42 converts the number of steps input from the pedometer 62 into digital data (binary data) and inputs it to the CPU 40. An inclination angle sensor 64 is connected to the A / D converter 44, and the A / D converter 44 converts the inclination angle (angle) input from the inclination angle sensor 64 into digital data (binary data). Input to the CPU 40. A microphone 66 is connected to the encoder 46, and the encoder 46 modulates an audio signal input from the microphone 66 into compressed audio data such as MP3 (hereinafter simply referred to as “audio data”) and inputs it to the CPU 40. Thus, the audio signal is compression-modulated because the capacity of the memory card 68 described later is taken into consideration. If the capacity of such a recording medium can be ignored, it may be simply converted into digital data.

  A pyroelectric sensor can be used as the non-contact sensor 48, and the CPU 40 turns on and off the microphone 66 in accordance with an input from the non-contact sensor 48. In this embodiment, when the nurse raises and lowers his / her hand twice in front of the non-contact sensor 48, that is, the pyroelectric sensor, the detection signal is input to the CPU 40. In response to this, the CPU 40 turns on the microphone 66, Thereafter, when the nurse raises and lowers his / her hand twice in front of the pyroelectric sensor, the microphone 66 is turned off. However, the CPU 40 is configured to turn off the microphone 66 even when a predetermined time (10 seconds) has elapsed without depending on the operation of the nurse. In this embodiment, a headset type microphone is used as the microphone 66 (see FIG. 4) and has directivity. This is to accurately input the voice of the nurse and protect the privacy of the patient.

  The interface 50 is an interface such as a LAN (wireless LAN) adapter, whereby the sensor unit 20 is directly connected to the network 14. However, by providing an interface such as RS232C or USB and using a cable, it is possible to connect to another terminal such as the server 12 or the terminal 16 so as to be able to communicate. A memory card 68 such as an MMC is detachably provided in the card slot 52. The memory card 68 stores data (step count data, inclination angle data, audio data) detected or acquired by the sensor unit 20. Further, the data stored in the memory card 68 can be moved (copied) to the other terminal by mounting the memory card 68 on the other terminal such as the server 12 or the terminal 16.

  The clock circuit 54 counts the date and time, and the CPU 40 adds the time data acquired from the clock circuit 54 to the data acquired from each sensor (62, 64 and 66) and the data of the nurse ID. And stored in the memory card 68. If necessary, the data to which the time data is attached is output to the server 12 and / or the terminal 16 connected to the network 14 via the interface 50.

  The DIP switch 56 is composed of, for example, 8 bits, and a numerical value can be set between 0 and 255 by switching on / off of each bit. This numerical value is identification information, that is, a nurse ID, and a different value is set for each sensor unit 20. The CPU 40 attaches the nurse ID as a label, and stores the data with the time data added to the memory card 68 or transfers it to the server 12.

  In this embodiment, the nurse ID is set using the DIP switch 56, but should not be limited to this. For example, a nurse ID can be stored in a memory card 68 described later. Further, instead of the DIP switch 56, a ROM or the like storing a nurse ID may be provided.

  The wireless transmitter 58 transmits the nurse ID set by the DIP switch 56 by radio waves (weak radio waves) at a predetermined frequency in accordance with instructions from the CPU 40. The wireless receiver 60 receives a weak radio wave transmitted from another sensor unit 20 existing in a wireless communicable range, demodulates it to a nurse ID, and inputs data about the demodulated nurse ID to the CPU 40.

  Here, as described above, the station 18 detects the nurse ID of the sensor unit 20 and transmits the detected nurse ID to the server 12 via the network 14. Also, identification information (station ID) is assigned to the station 18, and the station ID is detected by the sensor unit 20 as described above. Thus, for example, the station 18 can be configured by using some circuit components of the sensor unit 20. Specifically, the station 18 includes a CPU 40, an interface 50, a DIP switch 56, a wireless transmitter 58 and a wireless receiver 60.

  Note that a ROM that stores a station ID may be provided instead of the DIP switch 56 as in the case of the sensor unit 20.

  The sensor unit 20 configured as described above is attached to each nurse. For example, as shown in FIGS. 3 and 4, circuit components other than the non-contact sensor 48, the pedometer 62, the tilt angle sensor 64, and the microphone 66 are accommodated in a box (housing) 70, and the box 70 is a nurse's waist. Attached to (belt part). The non-contact sensor 48 and the tilt angle sensor 64 are housed in a pen-shaped case 72 so as to be inserted into a breast pocket of a nurse's clothes (white robe). However, in the drawings, the case 72 is shown outside the breast pocket for easy understanding. The pedometer 62 is attached to the nurse's lower back as in the box 70 described above. Furthermore, the microphone 66 is mounted on the nurse's head.

  Although not shown in FIG. 4, the non-contact sensor 48 is connected to the CPU 40 in the box 70 using a connection line, and the pedometer 62, the inclination angle sensor 64, and the microphone 66 are each connected to the box using a connection line. The A / D converter 42, the A / D converter 44, and the encoder 46 in 70 are electrically connected. However, you may make it connect by short-distance radio | wireless like Bluetooth, without using a connection line. That is, it is only necessary to be electrically connected.

  Also, as shown in FIG. 4, the nurse stores and carries (carries) a portable computer (PDA in this embodiment) 80 in the front pocket of a lab coat, for example. Although not shown, the PDA 80 is configured to be connectable to the network 14 shown in FIG. 1 by a wireless LAN. Since the PDA 80 is already known, a detailed description of its configuration and operation will be omitted. In the drawing, it is shown outside the front pocket for easy understanding.

  For example, when the main power supply of the sensor unit 20 is turned on by the operation of the nurse, the CPU 40 detects the input from the tilt angle sensor 64 every predetermined time (in this embodiment, 500 ms). As described above, the microphone 66 is controlled to be turned on / off according to the input of the non-contact sensor 48. Further, the CPU 40 validates the input from the pedometer 62 when the microphone 66 is turned off.

  Therefore, the CPU 40 uses the step count data (step count data) measured by the pedometer 62, the angle data (angle data) measured by the tilt angle sensor 64, and the voice data (voice data) detected by the microphone 66. The data is classified and stored in the memory card 68, and transmitted to another terminal (in this embodiment, the server 12) by wireless LAN. Further, when recording the step count data, the angle data, and the audio data, the CPU 40 acquires time data about the acquisition start time and the acquisition end time of each data from the clock circuit 54, and attaches the time data to each data. It is like that.

  For example, when nurse n visits patient k's patient, nurse n remarks that he is going to visit patient k's (patient). Data on each of the number of steps and the angle (tilt angle) when the upper body is tilted while reciprocating the hospital room is recorded. In addition, time data of the start time and end time of nursing work is added to this data.

  Such data on the voice, the number of steps and the inclination angle and the time data attached thereto (hereinafter, these may be collectively referred to as “measurement data”) are all about the daily work of the nurse n. To be acquired. As described above, the sensor unit 20 is connected by the wireless LAN (network 14). Therefore, the measurement data (business measurement data) for the daily work acquired as described above is transmitted from the sensor unit 20 to the terminal. 16 can be transmitted (transferred). If the memory card 68 is mounted on the terminal 16, the business measurement data can be directly input to the terminal 16.

  The terminal 16 receives the transferred work measurement data, and the work measurement data is uploaded to the server 12 by the operation of a nurse, for example. The server 12 creates nurse data excluding the accident / incident based on the uploaded work measurement data. That is, the nurse is specified, and thereafter, the items of the nurse data excluding the accident / incident are specified (determined) from each measurement data and recorded.

  However, the business measurement data may be directly uploaded from the sensor unit 20 to the server 12 via the network 14.

  For example, the nurse can easily specify on the server 12 side based on the nurse ID attached to the work measurement data transferred from the sensor unit 20.

  The voice data included in the measurement data is subjected to voice recognition processing, and the patient name and the nursing work performed on the patient are specified from the recognized voice. That is, the server 12 has a voice recognition function, and also has a dictionary for voice recognition, that is, a memory (for example, a hard disk or a ROM) that records voice data corresponding to a plurality of utterances (voices). . However, the voice data of the nurse is recorded in the nurse DB 26 as an example.

  In this embodiment, the plurality of utterances are utterances related to nursing work, and the voice data corresponding to the utterances related to the plurality of nursing work is recorded (recorded) when the nurse inputs the voice in advance. However, in order to recognize each utterance efficiently, a dictionary is provided for each nurse.

  Therefore, the server 12 compares the voice data of the nurse's voice with the voice data recorded in the memory by the DP matching method or the HMM (Hidden Markov Model) method, and recognizes (identifies) the contents spoken by the nurse. ) Specifically, when the utterance “I will go to urinalysis of patient k” is recognized, the patient name “patient k” and the content of nursing work “urinalysis” for the patient can be specified. Further, when the utterance “The urinalysis of the patient k has been completed” is recognized, it is possible to recognize that the content of the nursing work “urinalysis” for the patient has been completed.

  However, since it is necessary to specify the patient name and the contents of nursing work accurately, the highest (first place) utterance and the second highest degree of approximation (second) from the comparison result by the DP matching method etc. The second utterance is extracted, and the validity (reliability) of recognition of the first utterance is determined based on the first utterance and the second utterance.

  That is, the patient name (first patient name) identified from the first utterance and the contents of the nursing work (first nursing work) are extracted, and the patient name (second patient utterance) identified from the second utterance ( Each of (second patient name) and contents of nursing work (second nursing work) are extracted. Next, the degree of approximation (distance value) between the first patient name and the second patient name and the distance value between the first nursing service and the second nursing service are calculated. Here, the distance value refers to the degree to which the pronunciation of a word (word) is approximate. For example, “Ito” and “Kito” have a high degree of approximation (small distance value), but “Ito” and “Tanaka” have a low degree of approximation (distance) Value is large).

  Therefore, when the distance value between patient names and the distance value between nursing duties are smaller than a preset threshold (first threshold), the first utterance and the second utterance are approximated. Therefore, it can be said that the reliability of the recognition result is low. On the other hand, when the distance value between patient names and the distance value between nursing duties are larger than the first threshold, the first utterance and the second utterance are not approximated. It can be said that the degree is high.

  In this embodiment, when both the distance value between patient names and the distance value between nursing services are small, the first utterance and the second utterance are approximated, and the recognition result Although the reliability is considered to be low, the reliability of the recognition result may be considered to be low when one of the distance values is small.

  When the reliability of the recognition result is high, the patient name and the content of nursing work included in the first utterance are extracted (adopted) and used to create nurse data. On the other hand, when the reliability of the recognition result is low, the patient name and the content of the nursing work included in the first utterance are excluded (rejected) and manually input by the administrator of the server 12. That is, the manager refers to the daily work report and the like, and inputs the patient name and nursing work using an input device (keyboard and computer mouse) not shown.

  However, the patient name can be specified based on the location information data without using voice recognition. For example, the station 18 is assigned to each patient (the bed), and the time corresponding to the time data added to the voice data and the nurse ID added to the work measurement data are read out. Next, the location information data at the time is read from the location information DB 34, and the station ID that detects the nurse ID is specified. Then, the patient name is specified from the installation position of the station 18 to which the station ID is assigned.

  Further, the data to be entered in the items of the start time and end time included in the nurse data is determined from the time data added to the voice data. That is, the recording start time included in the time data is described in the start time column, and a value (time) obtained by subtracting the recording start time from the recording end time is described in the end time column.

  Table 2 shows the items included in typical accident and incident reports.

  The report shown in Table 2 is written by the staff involved in the incident, the incident, or the staff who discovered it. There are two types of report formats. One is a selection from a list such as date and time or a standard input. The other is a free description.

For example, “reporter information” is selected from “party” or “discoverer”. “Discoverer: occupation” such as nurse or doctor is also selected from the list or entered in the same format.
On the other hand, “occurrence information: detailed information”, “occurrence information: response after occurrence”, “occurrence information: influence on patient”, “occurrence information: cause and countermeasure” are written in free description.

  For example, “Dose” is written in “Occurrence information: Type”, and “… Instructions are written apart in“ Occurrence information: Occurrence status ”. "It has been ..." is written. In the “occurrence information: causes and countermeasures”, important information obtained from the analysis is written. For example, “It was a busy morning night shift. The morning was also being chased by temperature measurement, treatment, and care…”.

  For future analysis, accident and incident reports should contain as much information as possible, and self-analysis of accidents and incidents will be written. Therefore, the free description part is important for analysis.

  In this way, the report includes free description, but it is highly likely that this free description includes environmental information and unpredictable factors. On the other hand, in the analysis, if only items such as options are used, environmental information or information that cannot be predicted will be lacking. It is important to obtain an explicit result on a computer by analyzing items with only such options. However, the results are inadequate from a nursing risk management perspective.

  Therefore, it is important to analyze the free description of the accident / incident report as shown in Table 2, but it is difficult to obtain a complicated relationship from such simple data. Furthermore, because reports lack a lot of information, such as the environment and relationships between nurses, it is very difficult to get complex results from reports alone.

  Accident and incident reports are usually written to prevent the same or similar accidents and incidents. Therefore, as shown in Table 2 above, it is written accurately so that other staff can understand why, when, by whom, and how the accident or incident occurred.

  Thus, the report contains important information about accidents and incidents. Therefore, risk managers, when they receive the report, scrutinize the content to find critical points in accidents and incidents and show how they can prevent such nursing accidents in the future.

  For example, “Nuclear medicine tests are often cancelled, so we prepared only by sucking distilled water into the syringe, and Diamox was going to dissolve just before, out of the box and placed on the side of the syringe. As we proceeded, we assumed that it was dissolved and gave it to the doctor. ”Upon receiving the accident and incident report, the person who conducts risk management said,“ Suspension of work is a cause of error. "Basically, let's make a procedure that does not require interruptions." In addition, point out or suggest missing information that is the cause of inability to understand and understand the situation.

  Thus, basically, accident and incident reports are analyzed by human experienced persons. However, it would be even better if it could be analyzed automatically by a computer. This is because high-speed analysis can be performed using a computer. This is particularly effective when the data contained in the report is very large. In addition, when analyzed using a computer, there is no room for subjectivity, so data can be analyzed naturally and objectively.

  By analyzing the description in the free description, environmental factors related to nursing accidents and incidents can be discovered. However, it is not easy to analyze a free description as it is. To simplify it, text mining techniques can be applied. Furthermore, methods such as summarization and focus extraction are necessary to obtain important information for nursing risk management. In addition, nursing risk management needs data that includes both explicit and implicit (potential) information that has some connection to the environment. It is also important to look for potential relationships from this data that can cause unintentional and serious accidents.

  As pointed out above, for effective nursing risk management, it is necessary to acquire data that includes both explicit and implicit (potential) information that is somehow related to the environment. However, it is very difficult for individuals to read all environmental factors from accident and incident reports. This is because not all environmental information is written in the report. Therefore, in addition to accident and incident reports, an automated environmental monitoring system is necessary to capture all environmental factors.

  Therefore, in this embodiment, the nursing activity monitoring system 10 shown in FIGS. 1 to 4 is used. Using this system 10, it was proposed to monitor and report a nurse's activity workflow on behalf of a nurse in real time. That is, the nurse's workflow is automatically monitored and recorded. As a result, various types of data regarding nursing activities are acquired.

  For example, from the number of feet and the inclination angle of the body, it is possible to predict work categories such as intuitive care at the bedside, assisting patient walking, and creating a document. In addition, if you use a video monitor system, you can see the above category predictions and discover that the nurse is unaware. In the future, the integrated nursing activity system will be expanded and can be used as a real-time monitoring and alarm system for nursing activities.

  While the nurse is active, the monitor system records the nurse's behavior and analyzes and evaluates the nurse's behavior data to predict possible mistakes. When the system detects a possible nursing error, the system alerts the nurses involved.

  As described above, data relating to various types of nursing activities can be acquired by the systems shown in FIGS.

  Here, as an example, Table 3 shows the results of data mining of nurse conversations. In the experiment, “C4.5” was adopted. “C4.5” is disclosed by Quinlan J. R., for example, in Programs for Machine Learning, Morgan Kaufman, 1993. A decision tree can be generated from a data set and used as a data mining tool. The advantage of generating a decision tree is that it can be modeled as a propositional rule or a set of causal relationships. In other words, from the decision tree, we can find a set of generalized rules and causal relationships that can be used to make inferences about nurse behavior.

  The following is a part of a decision tree generated from a nurse's conversation by “C4.5”. A nurse working with a microphone is supposed to verbally record all actions. The microphone 66 (FIGS. 3 and 4) basically records only the voice of the nurse wearing the microphone.

  The data includes time information. However, since “C4.5” cannot handle time information, the notation of time is slightly changed. In other words, the start time and end time were converted into the start time (XX hour) and duration (XX minutes). For example, if the job starts at 9:46 and ends at 9:53, the start time is “9” and the duration is “7”.

  This result shows the time required for typical nursing work and typical nursing activities. However, since the above result is generated only from the nurse's own conversation data, no problem or environmental information is included.

  In the case of trouble, it seems very difficult to report the behavior calmly. On the other hand, in the case of a serious accident, it will be silent. In addition, some accidents occur unconsciously.

  Therefore, it seems difficult to find the causality related to a nursing accident from a decision tree generated only from nurse's conversation data.

  Thus, it is necessary to obtain nursing accidents, information on incidents, environmental information, and the like from sources such as accidents, incident reports, and videos.

  Here, the results of data mining from the accident / incident report and the data mining results from the data recorded by the system 10 of the embodiment will be examined.

  As explained earlier, the results of data mining from the options included in accident and incident reports are very simple.

  Looking at the decision tree (Table 1), it is possible to obtain only a simple relationship such as “the cause of the wrong drug is insufficient explanation to the patient” or “the cause of the wrong drug is insufficient observation”. . This is because the analyzed data is classified into simple categories before that. In addition, only items that can be predicted in advance are included. Therefore, the result is only a natural result. Actually, “Others” is included in the options, but it is meaningless information for analysis by a computer.

By analyzing the description included in the free description column, the environmental reasons for nursing accidents and incidents can be found, but in order to perform the analysis, techniques such as summarization and focus extraction are required. However, since I haven't noticed everything over time, there seems to be something missing in the free description. In addition, because the report is not written in real time, the memory about an event may have disappeared. The report also contains information that cannot be described. One example is common knowledge that does not need to be written, and other tacit knowledge that cannot be written. Therefore, it seems difficult to obtain the full reason for nursing accidents and incidents even if summarization techniques such as focus extraction are applied. Thus, techniques other than extraction are required. Thus, there is a considerable lack of factors in the data mining results from the monitored data. Further, since the monitored data is not taken in association with an accident or an incident, it is necessary to associate the data with an accident or an incident. In addition, the data must be associated with environmental information and other relevant information. Because all data is taken almost independently. After recording, it is necessary to look for their relationship by some means.

  For example, by using information generated by an RFID tag (not shown), the position information of the nurse can be obtained, and the interaction between the nurses can be estimated. And by using time stamp information, you can observe the situation when the interaction occurs by looking at the image taken in the video.

  These data should be linked to make it easier to find relationships. Combined or linked datasets can be powerful data in obtaining relationships and causality of nursing incidents and incidents. In addition, it is very important to combine the monitored data with nursing accident and incident reports.

  As mentioned above, nursing accidents and incident reports sometimes lack an environmental aspect. If you can find a link from the report to the monitored data, you can supplement the missing information.

  As already explained, data mining from a single resource is not enough. Because every resource data lacks something. Therefore, in order to obtain appropriate rules and relationships, it is necessary to handle integrated data. The hospital records nursing accidents and incident reports when an accident or incident occurs. The report includes a human perspective, which is useful when analyzing accidents and incidents. However, because the report is written by the nurse and her colleagues, things that are well-known may have been omitted, forgotten or even not noticed.

  Thus, it is important to supplement the missing information from the report.

  Therefore, by combining the integrated nursing activity monitoring system 10 shown in FIGS. 1 and 2 with the nursing accident and incident report, a more detailed and accurate relationship between the accident and incident, the nurse and the environment is generated. is there. When a certain part of an accident or incident is missing, the nursing activity monitor system is referred to as shown in FIG. 5A to supplement (or interpolate) the missing information. Data mining is performed especially on the free description part of the accident / incident report having information supplemented in this way.

  For example, in the case of an incident report that says, “After taking a Kadian capsule, an empty sheet of narcotics was left on the meal tray and left behind, and the sheet became unknown.” In this case, the information of “who” Although it is missing, information such as the video around the time and location of the report is examined, the person who performed the above act (empty seat left), the process up to this accident, the environmental situation, etc., Correct the defect in the incident report. As a result, this deficiency in the incident report is eliminated, and more accurate data mining becomes possible.

  The complemented (interpolated) data is analyzed by a data mining technique. By using this integrated data mining system, nursing risk management can be performed in consideration of factors such as environmental factors and human factors.

  If the nursing activity monitoring system 10 lacks or does not understand any accident or incident information, the system 10 refers to the accident / incident report as shown in FIG. Complement (or interpolate) the information that is present. Data of the integrated nursing activity monitoring system 10 having information supplemented in this way is data mined.

  The SHELL model proposed by Edwards, Hawkins, etc. is known as a method for analyzing human mistakes in consideration of the relationship between two or more people and the environment and interaction.

  The system in this example is similar in concept to the SHELL model, but the SHELL model does not assume that there is any missing information, whereas in this example it complements it when information is missing. It is completely different in that it does.

  In the integrated nursing activity monitor system 10 of FIG. 1, images and images are included as data. Therefore, if data mining is performed automatically by a computer, a symbol grounding problem will occur. The symbol grounding problem is quite difficult and is still being discussed at AI.

  However, it seems that the symbol grounding problem can be solved by interacting with a computer. Because if the computer points out the problem, and they are correct, you can find the important point by checking the video monitor or looking at other devices. In other words, it is not necessary to convert images and symbols into symbols.

  In the example, conversation data was used for analysis. Therefore, conversation information such as filler, intonation, and conversation length is missing. Such information may need to be analyzed in case of trouble or panic.

  In addition, “incident” is usually used as a term relative to an accident (medical accident), but it is a mistake or wrong business action related to medical nursing work, and is reported by an incident report that is a report from a nurse etc. All that is done is called an incident. Therefore, this incident does not necessarily have a negative effect on the patient, that is, a medical accident. In contrast, an “accident” is an accident caused by medical treatment (nursing) by a doctor (nurse).

FIG. 1 is an illustrative view showing one example of a configuration of a medical accident information analysis system of the present invention. FIG. 2 is an illustrative view showing an electrical configuration of the server shown in FIG. FIG. 3 is an illustrative view showing an example of an electrical configuration of a sensor unit worn by a nurse. FIG. 4 is an illustrative view showing a state in which the sensor unit shown in FIG. 3 is attached to a nurse. FIG. 5 is an illustrative view showing a state in which information is supplemented according to the embodiment of the present invention. FIG. 5 (A) shows a state in which data missing is performed by supplementing the information on accident and incident reports from the integrated nursing activity monitoring system. FIG. 5B shows a state in which data mining is performed by complementing the lack information of the integrated nursing activity monitoring system from the accident / incident report.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Integrated nursing activity monitoring system 12 ... Server 16 ... Terminal 18 ... Station 20 ... Sensor unit 62 ... Pedometer 64 ... Inclination angle sensor 66 ... Microphone 80 ... PDA

Claims (3)

An accident / incident report analysis system that analyzes nursing accidents and incident reports using data mining,
An accident / incident report analysis system which supplements missing information with reference to a nursing activity monitoring system when analyzing the free description of the report.   The accident / incident report analysis system according to claim 1, wherein information lacking in the nursing activity monitoring system is supplemented with reference to the report.   The accident / incident report analysis system according to claim 1 or 2, wherein information of the nursing activity monitoring system is analyzed by data mining.

Images