WO2019008798A1 - Dispositif de prédiction d'apparition de maladie, procédé de prédiction d'apparition de maladie et programme - Google Patents

Dispositif de prédiction d'apparition de maladie, procédé de prédiction d'apparition de maladie et programme Download PDF

Info

Publication number
WO2019008798A1
WO2019008798A1 PCT/JP2018/000023 JP2018000023W WO2019008798A1 WO 2019008798 A1 WO2019008798 A1 WO 2019008798A1 JP 2018000023 W JP2018000023 W JP 2018000023W WO 2019008798 A1 WO2019008798 A1 WO 2019008798A1
Authority
WO
WIPO (PCT)
Prior art keywords
onset
disease
biological data
prediction
information
Prior art date
Application number
PCT/JP2018/000023
Other languages
English (en)
Inventor
Daria Antonia BUNU
Takashi Okada
Junko Yano
Original Assignee
Ntt Data Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ntt Data Corporation filed Critical Ntt Data Corporation
Publication of WO2019008798A1 publication Critical patent/WO2019008798A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/41—Detecting, measuring or recording for evaluating the immune or lymphatic systems
    • A61B5/412—Detecting or monitoring sepsis
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271—Specific aspects of physiological measurement analysis
    • A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/67—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for remote operation
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A61B5/024—Detecting, measuring or recording pulse rate or heart rate
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/08—Detecting, measuring or recording devices for evaluating the respiratory organs

Definitions

  • the present disclosure relates to a disease onset prediction device, a disease onset prediction method, and a program.
  • Patent Literature 1 discloses a disease prediction device for finding a relationship between 2 parameters indicating biological data of the patient, comparing the found relationship to a statistical value, and analyzing signs of disease.
  • Patent Literature 2 discloses a method for modeling health status of a person by modeling biological data of a person and analyzing a difference between a hypothesized value obtained from the model and an actual measured value.
  • Patent Literature 1 only detects the occurrence of a change in the patient on the basis of a comparison with a statistical value. That is to say, the technology disclosed in Patent Literature 1 cannot detect with good accuracy signs occurring prior to the onset of a specific disease. Further, the method disclosed in Patent Literature 2 merely specifies whether the actual measured value is that of the healthy state, and this method does not predict the onset of a specific disease.
  • the present disclosure is developed in consideration of the aforementioned circumstances, and an objective of the present disclosure is to provide a disease onset prediction device, a disease onset prediction method, and a program that are capable of good accuracy in prediction of the onset of a specific disease.
  • a disease onset prediction device includes: biological data acquisition means for acquiring biological data of a prediction target and prediction model construction biological data; onset information acquisition means for acquiring onset information indicating whether a patient corresponding to the prediction model construction biological data had an onset of a disease; sample extraction means for extracting, when the onset information indicates that the patient had the onset of the disease: (i) biological data to form positive samples from, among the prediction model construction biological data, wherein the biological data was measured from a standard time period prior to an onset time of the disease until the onset time of the disease, and (ii) biological data to form negative samples from the biological data wherein the biological data was measured earlier than the standard time period before the onset time; prediction model construction means for constructing a prediction model by using the positive samples as training data for the onset of the disease within the standard time period, and using the negative samples as training data for absence of the onset of the disease within the standard time period; and onset prediction means for, based
  • the sample extraction means may form the positive samples by multiple overlapping extractions of a prescribed time period portion from the prediction model construction biological data when the onset information indicates that the patient had the onset of the disease, wherein the biological data corresponds to a prescribed time period and was measured from a total time period prior to the onset time to the onset time, the total time period being a sum of the standard time period and the prescribed time period.
  • the disease onset prediction device may be provided with action information acquisition means for acquiring action information indicating content of an action performed with respect to the patient corresponding to the prediction model construction biological data.
  • the sample extraction means performs the extraction to extract the biological data to form the positive samples or the negative samples from the prediction model construction biological data preferentially based on the action information.
  • the sample extraction means when the action information indicates an action for preventing the onset of the disease, may hypothesize the onset of the disease for the patient corresponding to the prediction model construction biological data and extracts the biological data forming the positive samples in the same manner as when the onset data indicates that the patient had the onset of the disease.
  • the action information acquisition means may further acquire prediction target action information indicating content of an action performed on the patient corresponding to the biological data of the prediction target predicted by the onset prediction means to have the onset of the disease; the biological data acquisition means may newly acquire, as the prediction model construction biological data, data corresponding to the biological data of the prediction target of the patient subject to the performed action; and when the prediction target action information does not indicate a prior action for preventing the onset of the disease, the sample extraction means may extract, with priority over other biological data, the newly acquired prediction model construction biological data as the negative samples.
  • the prediction model construction means may, based on the biological data of the prediction target, set a condition to output information on caution indicating that the action for preventing the onset of the disease is to be performed; and the disease onset prediction device may further comprise information on caution output means for, when the prediction target action information does not indicate the prior action for preventing onset of the disease, outputting the information on caution in accordance with the condition set by the prediction model construction means.
  • Alert information output means may be further provided for, upon the onset prediction means predicting the onset of the disease, based on the biological data of the prediction target, classifying, by a predetermined method, a component indicating a reason of the disease onset prediction and outputting the component as alert information.
  • a disease onset prediction method includes: acquiring biological data of a prediction target and prediction model construction biological data; acquiring onset information indicating whether a patient corresponding to the prediction model construction biological data had an onset of a disease; extracting when the onset information indicates that the patient had the onset of the disease: (i) biological data to form positive samples from, among the prediction model construction biological data, wherein the biological data was measured from a standard time period prior to an onset time of the disease until the onset time of the disease, and (ii) biological data to form negative samples from the biological data wherein the biological data was measured earlier than the standard time period before the onset time; constructing a prediction model by using the positive samples as training data for the onset of the disease within the standard time period, and using the negative samples as training data for absence of the onset of the disease within the standard time period; and based on the biological data of the prediction target and the prediction model, predicting whether the onset of the disease will occur in
  • a program may cause a computer to execute: acquiring biological data of a prediction target and prediction model construction biological data; acquiring onset information indicating whether a patient corresponding to the prediction model construction biological data had an onset of a disease; extracting when the onset information indicates that the patient had the onset of the disease: (i) biological data to form positive samples from, among the prediction model construction biological data, wherein the biological data was measured from a standard time period prior to an onset time of the disease until the onset time of the disease, and (ii) biological data to form negative samples from the biological data wherein the biological data was measured earlier than the standard time period before the onset time; constructing a prediction model by using the positive samples as training data for the onset of the disease within the standard time period, and using the negative samples as training data for absence of the onset of the disease within the standard time period; and based on the biological data of the prediction target and the prediction model, predicting whether the onset of the disease will
  • a disease onset prediction device capable of good accuracy in prediction beforehand of an onset of a specific disease.
  • FIG. 1 is a drawing schematically illustrating an alert system according to an embodiment of the present disclosure.
  • FIG. 2 is a functional block diagram of the alert system according to the embodiment of the present disclosure.
  • FIG. 3 is a drawing illustrating a sample extraction method of the alert system according to the embodiment of the present disclosure.
  • FIG. 4 is a table indicating one example of a sample extracted by the alert system according to the embodiment of the present disclosure.
  • FIG. 5 is a hardware configuration diagram of the alert system according to the embodiment of the present disclosure.
  • FIG. 6 is a flowchart of prediction model construction processing of the alert system according to the embodiment of the present disclosure.
  • FIG. 7 is a table indicating one example of a sample of principal components extracted in alert reason classification condition setting processing of the alert system according to the embodiment of the present disclosure.
  • FIG. 8 is a table indicating one example of principal component scores imparted in the alert reason classification condition setting processing of the alert system according to the embodiment of the present disclosure.
  • FIG. 9 is a table indicating one example of samples imparted classifications in the alert reason classification condition setting processing of the alert system according to the embodiment of the present disclosure.
  • FIG. 10 is a flowchart of prediction model reconstruction processing of the alert system according to the embodiment of the present disclosure.
  • FIG. 11 is a table indicating one example of setting values in the prediction model reconstruction processing of the alert system according to the embodiment of the present disclosure.
  • FIG. 12 is a flowchart of alert processing of the alert system according to the embodiment of the present disclosure.
  • FIG. 13 is a drawing indicating one example of an output screen of alert information displayed on a portable terminal according to the embodiment of the present disclosure.
  • FIG. 14 is a drawing indicating one example of an action selection screen displayed on the portable terminal according to the embodiment of the present disclosure.
  • FIG. 15A is a graph indicating one example for description of a result of the alert system according to the embodiment of the present disclosure.
  • FIG. 15B is a graph indicating another example for description of the result of the alert system according to the embodiment of the present disclosure.
  • Embodiment 1 An embodiment applying a disease onset prediction device of the present disclosure to an alert system for notification of a prediction of a disease of a patient is described below with reference to drawings.
  • the alert system 1 of the present embodiment is a disease onset prediction device of the present disclosure, and is a system for prediction of an onset of the disease of the patient within a standard time period and for sending notification to a physician in charge of the patient.
  • the alert system 1 is set beforehand for one specific disease or multiple specific diseases. Then for each such set disease the alert system 1 predicts whether the target patient will have an onset of the disease. Upon prediction of the onset of the disease, the alert system 1 notifies the physician of the prediction.
  • a manager of the alert system 1 sets the alert system 1 to a standard time period determined beforehand for each disease, such as 3 hours for blood poisoning or 5 hours for cardiovascular disease.
  • the alert system 1 receives in real time via a hospital network 5 vital signs data of the patient measured by a below-described medical instrument 2.
  • the alert system 1 analyses the received vital signs data and predicts whether the patient will have the onset of disease.
  • the alert system 1 analyses the predicted vital signs data and generates alert information.
  • the alert system 1 transmits the generated alarm information to a below-described portable terminal 4 via the hospital network 5.
  • the physician acquires from the portable terminal 4 handing information, indicating content of operating the below-described portable terminal 4, so that the alert system 1 reflect the content in a below-described prediction model.
  • the alert system 1 communicably interconnects the medical instrument 2, the hospital information system 3, and the portable terminal 4 via the hospital network 5 as illustrated in FIG. 1.
  • the medical instrument 2 is an instrument that measures the vital signs data of the patient.
  • the medical instrument 2 is typically an instrument that measures a vital sign of the patient in an intensive-care unit.
  • the medical instrument 2 may also be any instrument that measures vital signs data of the patient, such as an artificial respirator or an infusion pump.
  • vital signs data may be other biological data.
  • the expression "biological data of the patient” refers to information mainly quantifying information related to health status of the patient.
  • the biological data includes data of so-called vital signs such as blood pressure, pulse rate, respiration rate, body temperature, and the like, as well as data such as blood platelet count, bilirubin, mean arterial blood pressure, urinary volume, and the like.
  • the medical instrument 2 measures the vital signs data periodically every second, for example. Then the medical instrument 2 transmits the measured vital signs data to the alert system 1 and the hospital information system 3 via the hospital network 5.
  • the hospital information system 3 is an information system set up in the hospital and includes subsystems such as an electronic medical record system, a checkup information system, a pharmaceutical agent system, and a patient information system.
  • the hospital information system 3 stores basic information, administered pharmaceutical agents, an electronic medical record, and the like of the patient. Further, the hospital information system 3 receives the real time vital signs data of the patient from the medical instrument 2 via the hospital network 5, and stores the received vital signs data, together with a measurement time, in an internal storage device.
  • the hospital information system 3 transmits to the alert system 1 via the hospital network 5 information such as the basic information, the administered pharmaceutical agents, the electronic medical record, and vital signs data stored in the past, and the like of the patient.
  • medical staff of the hospital can access the hospital information system 3 via various non-illustrated equipment and can refer to the stored information. Further, the hospital information system 3 receives the operating information of the physician from the below-described portable terminal 4. The medical staff can thus visually check the content of the operation by the physician.
  • the portable terminal 4 is a terminal, such as a smart phone, hospital-clinic portable terminal, tablet, personal digital assistant (PDA), or cellphone, carried by the physician.
  • the portable terminal 4 displays on a screen the alert information received from the alert system 1.
  • the portable terminal 4 thus allows the physician to visually check the alert information.
  • the portable terminal 4 is operated by the physician. Then the portable terminal 4 transmits the operation content as operating information to the alert system 1 and the hospital information system 3 via the hospital network 5.
  • the medical staff acquires the operating information of the physician via the hospital information system 3.
  • the alert system acquires the operating information of the physician and reflects the operating information in updated content of the below-described prediction model.
  • the alert system 1 functions as an operational unit for performing a below-described prediction model construction processing, prediction model reconstruction processing, and alert processing.
  • the alert system 1 as the operational unit performing the aforementioned various types of processing, includes a vital signs data acquirer 11, an onset information acquirer 12, a sample extractor 13, a prediction model constructor 14, an onset predictor 15, an alert information outputter 16, and an action information acquirer 17.
  • CPU central processing unit
  • the alert system 1 is described prior to operation and during operation.
  • operation refers to the stage of actual use of the alert system 1 for the patient which is the prediction target.
  • prior to operation means at the preparatory stage prior to actual use.
  • the vital signs data acquirer 11 sends a request via the hospital network 5 to the hospital information system 3 for the vital signs data acquired in the past from multiple patients, and the vital signs data acquirer 11 receives the vital signs data in reply to the request. Further, in order to predict disease onset during operation, the vital signs data acquirer 11 receives via the hospital network 5 the vital signs data of the patient, who is the prediction target, transmitted in real time from the medical instrument 2.
  • the onset information acquirer 12 acquires onset information indicating whether the patient corresponding to the past vital signs data had an onset of the disease.
  • the alert system 1 stores beforehand a standard for determination from the vital signs data whether the patients had the onset of a specific disease. Then the onset information acquirer 12 refers to the standard and determines whether there is the onset of the specific disease. Further, rather than the alert system 1 determining the onset information from the vital signs data, onset information determined by medical examination by the medical staff (mainly physicians) may be input via equipment to the hospital information system 3.
  • the hospital information system 3 transmits the onset information via the hospital network 5 to the alert system 1.
  • the onset information acquirer 12 may acquire the onset information by receiving the onset information via the hospital network 5 from the hospital information system 3.
  • the onset information includes information such as an ID of the patient having the onset of the disease, a name of the acquired disease, and date-time of the onset.
  • the sample extractor 13 extracts the vital signs data to form samples for machine learning prior to operation or during operation.
  • the alert system 1 uses the extracted samples as training data for machine learning in order to construct and update the prediction model used for prediction of the onset of the disease.
  • the alert system 1 takes the samples for teaching the onset of the disease within the standard time period to be the positive samples, and takes the samples for teaching absence of the onset of the disease within the standard time period to be the negative samples.
  • the sample extractor 13 extracts the positive samples and the negative samples separately as vital signs data from the past vital signs data. Further, this standard time period is called the "alert target time period".
  • this standard time period is set to 180 minutes.
  • the sample extractor 13 determines, on the basis of the onset information acquired by the onset information acquirer 12, whether the past vital signs data acquired by the vital signs data acquirer 11 is the vital signs data of the patients having had the onset of the disease, or is the vital signs data of the patients having had no onset of the disease. Then the sample extractor 13 basically extracts the vital signs data to form the positive samples from the patients having had the onset of the disease. Further, the sample extractor 13 extracts the vital signs data to form the negative samples from the patients having had no onset of the disease.
  • the vital signs data of the patients having had the onset of the disease often does not indicate signs of disease onset at times prior to at least the alert target time period before the disease onset, and thus vital signs data of such times is taken to be extracted as the negative samples.
  • the sample extractor 13 extracts as subset samples the vital signs data for each fixed time period acquired from the vital signs data acquirer 11, the vital signs data being collected for prescribed time periods (for example, 30 minutes).
  • the sample extractor 13 calculates the vital signs data forming the extracted sample by use of a predetermined function.
  • This function may be a function for calculation from one type of vital signs data, such as for standard deviation, average, maximum, or minimum; or the function may be for calculation from multiple types of the vital signs data.
  • the result of the calculation for each function is associated with the status of the patient during the 30 minutes, and this associated result is termed a "characteristic amount”.
  • the aforementioned prescribed time period of 30 minutes is termed a "characteristic amount window time period”.
  • the method by which the sample extractor 13 extracts the sample from the vital signs data of the patient having had the onset of the disease is described in detail with reference to FIG. 3.
  • the alert system 1 extracts the vital signs data to form the positive samples from the vital signs data of the patients having had the onset of the disease, among sample candidates from a sample candidate (double-headed arrow 101 of FIG. 3) in which vital signs data is collected during a characteristic amount window time period taking the disease onset time as an end point to sample candidates (double-headed arrow 107 of FIG. 3) in which vital signs data is collected during a characteristic amount window time period taken prior to the aforementioned alert target time period as an end point predicting the onset of the disease of the patient.
  • negative samples are extracted among the sample candidates from sample candidates (double-headed arrow 108 of FIG. 3) in which vital signs data is collected for the characteristic amount window time period taking as an end point back the alert target time period plus the target-exclusion time period from the disease onset time, and from sample candidates (double-headed arrows 109, 110, and 111 of FIG. 3) in which vital signs data is collected for characteristic amount window time periods prior to the target-exclusion time period.
  • the sample extractor 13 repeats extraction from among these sample candidates sampled interval-by-interval back from the time of the onset of the disease for the patient.
  • the vital signs data capable of becoming the subjects of the positive samples is only the vital signs data of the limited time period of the patients having had the onset of the disease.
  • the acquirable number of the positive samples is relatively small in comparison to the acquirable number of the negative samples.
  • the number of negative samples is high, and thus the prediction of the alert system 1 may have a bias for predicting the absence of the onset of the disease.
  • the sample extractor 13 preferably extracts as candidates for the positive samples all the sample candidates of the range illustrated in FIG.
  • the sample extractor 13 preferably extracts for the negative sample random candidates from among sample candidates capable of being subjects, so that the number of the negative samples is similar to the number of the positive samples. However, when the number of the positive samples is high, the sample extractor 13 may extract the positive samples randomly from among the positive samples capable being subjects.
  • the vital signs data forming the negative samples acquired from the patient having had the onset of the disease in comparison particularly to the positive samples, are samples that are effective for characterizing a change occurring in the alert target time period prior to the disease onset.
  • the sample extractor 13 may be configured to prioritize extraction of the vital signs data of the patient having had the onset of the disease over the vital signs data of the patient lacking the onset of the disease.
  • the sample extractor 13 applies the aforementioned function to the vital signs data to form the positive samples acquired in this manner and calculates the characteristic amounts.
  • the calculated characteristic amounts are taken to be the positive samples.
  • the sample extractor 13 applies the aforementioned function to the vital signs data to form the negative samples and calculates the characteristic amounts. Then the calculated characteristic amounts are taken to be the negative samples.
  • Examples of the positive sample and the negative sample extracted by the sample extractor 13 are illustrated in FIG. 4.
  • the average blood pressure and the standard deviation of pulse rate are each a characteristic amount.
  • the sample extractor 13 extracts from the same patient multiple positive samples (records no. 1 to 3) and multiple negative samples (records no. 4 to 6).
  • the prediction model constructor 14 constructs the prediction model used for the alert system 1 to predict the onset of the disease in the patient. Specifically, for the positive samples and the negative samples extracted by the sample extractor 13, the prediction model constructor 14 selects the prediction model as one prediction model from among a support vector machine, a decision tree model, a logistic regression model, and the like. The evaluation indicator, evaluation method, or the like for selection of the prediction model is set beforehand as a prediction model construction parameter. The prediction model constructor 14, on the basis of the prediction model construction parameter, selects the prediction model and determines the model parameters passed on to the prediction model.
  • prediction scores are calculated for each of the samples on the basis of the model parameters passed on to the prediction model, that is, parameters such as the rules, equations, coefficients, and the like of the selected prediction model.
  • the prediction model constructor 14 in this manner constructs the prediction model. Further, the prediction model constructor 14 may construct a different prediction model for each disease. Further, the support vector machine, the decision tree model, the logistic regression model, and the like are widely known prediction models, and thus detailed description of such models is omitted here.
  • the prediction model constructor 14 determines an alert output threshold.
  • the alert output threshold is a threshold for determination of whether a prediction score (corresponding to a probability of the onset of the disease within the alert target time period) calculated by application of the prediction model by a below-described onset predictor 15 is a score such that the alert is to be output.
  • the prediction model constructor 14 may refer to previously set parameters to determine the alert output threshold prior to operation, or may determine the threshold as a central value of the prediction scores. Alternatively, the prediction model constructor 14 may set the threshold on the basis of business requirements such that the alert notification is issued less than once per 1 hour. Further, the prediction model constructor 14 acquires the output results of the alert via the hospital network 5 from the hospital information system 3 during operation and thus reconstructs the prediction model.
  • the prediction model constructor 14 reconstructs the prediction model on the basis of samples such as cases, referred to as “false positive samples”, in which the disease onset does not occur despite issuance of the alert, and cases, referred to as “false negative samples”, having the disease onset despite absence of sending of the alert. Then the prediction model constructor 14 updates the alert output threshold in response to the prediction score at the time of such prediction model reconstruction. Alternatively, performing of the updating of the prediction model constructor 14 may occur only for the alert output threshold.
  • the prediction model constructor 14 sets alert reason classification conditions.
  • the "alert reason classification condition” is a condition for use by a below-described alert information outputter 16 for, on the basis of a principal component that is the reason of generation of the alert, including in the alert information a classification of the characteristic amount.
  • the prediction model constructor 14 as the output of each characteristic amount of the positive samples, uses principal component analysis and extracts the principal components. Then the prediction model constructor 14 determines for each positive sample a score for each extracted principal component. Then the prediction model constructor 14, on the basis of the determined scores, performs clustering and sets the classified conditions as the alert reason classification conditions.
  • the term "principal component” refers to a component relating to the disease onset, such as breathing variation or blood pressure.
  • the “classification” characterizes the positive samples as having "values that tend to decline", “values that vary greatly”, and the like. Further, “clustering” is a basic data analysis method that divides a set of classification targets into sub-sets, and description of details of this method in the present specification is omitted. Further, the alert reason classification condition is not necessarily on the basis of principal component analysis.
  • the prediction model constructor 14 on the basis of the number and properties of the characteristic amounts, may create a visualization of characteristic amounts of the prediction target patient and averages of the characteristic amounts of the positive samples and the negative samples, and may specify the reason of generation of the alert.
  • the onset predictor 15 in accordance with the prediction model constructed by the prediction model constructor 14, analyses the vital signs data transmitted from the medical instrument 2 via the hospital network 5. Then the onset predictor 15 calculates the prediction score indicating the probability that the onset of the disease occurs for the target patient in the alert target time period.
  • the alert information outputter 16 prepares alert information on the basis of the alert reason classification condition set by the prediction model constructor 14. Then the alert information outputter 16 transmits the prepared alert information to the portable terminal 4 via the hospital network.
  • the alert information includes the principal component indicating the reason of the disease onset prediction and information indicating the assigned classification (see FIG. 13). Further, the characteristic amount, or the vital signs data corresponding to the characteristic amount, related to the principal component indicating the reason of the disease onset prediction may also be included.
  • the action information acquirer 17 receives the content (referred to hereinafter as the action information) of operating of the portable terminal 4 by the physician.
  • the term “action information” refers to a type of medical action selected from among choices on the portable terminal 4 and displayed as alert information such as "administer pharmaceutical", “continue monitoring", “no problem", and the like. Further, this action information is transmitted from the portable terminal 4 to the hospital information system 3 via the hospital network 5.
  • medical staff such as a nurse performs the medical action on the patient.
  • the content of the medical action may be any action as long as the medical staff, by accessing the hospital information system 3, can perceive the action information selected by the physician.
  • the content of the aforementioned choices can be generally described in the below manner.
  • the choice “administer pharmaceutical” means performing a medical action with respect to the patient.
  • the choice “continue monitoring” means continuing monitoring of the status of the patient.
  • the choice “no problem” means that there is no need for continuation of monitoring. Further, these choices are merely examples, and any content may be used as long as the objective and functions of the disclosure are satisfied.
  • the action information includes a patient ID identifying the patient and the date-time when the alert is generated. Then the action information acquirer 17 stores in the alert system 1 the action information acquired from the portable terminal 4. When the alert system 1 performs machine learning during operation, the alert system 1 acquires from the stored action information the action information for use in learning.
  • the condition on caution setter 18 specifies the patient for which the action information acquired by the action information acquirer 17 is "no problem". Then among the onset information acquired by the onset information acquirer 12, the condition on caution setter 18 analyses the onset information of this specified patient. Then the condition on caution setter 18 sets a condition on caution for output of the information on caution alerting the physician of the selection of the "no problem” action. Specifically, among the patients which are targets of the "no problem” action, the condition on caution setter 18 sets the condition on caution by performing comparison between the vital signs data of patients having the disease onset thereafter and the vital signs data of patients having no disease onset, and then performing machine learning such as decision tree analysis. In this case, the condition on caution setter 18, rather than by machine learning, may set the condition by a rule base.
  • condition on caution setter 18 determines an information on caution threshold.
  • the information on caution threshold is a threshold for a below-described information on caution outputter 19 to determine whether a score on caution calculated by application of the conditions on caution is a score to output the information on caution.
  • the condition on caution setter 18 may determine the information on caution threshold by referring to parameters set beforehand, or may set the threshold to a central value of the scores on caution.
  • the information on caution outputter 19 calculates the score on caution on the basis of the condition on caution set by the condition on caution setter 18. Then in the case in which the calculated score on caution is greater than or equal to the information on caution threshold, the information on caution outputter 19 outputs the information on caution to the portable terminal 4 via the hospital network 5. Further, the information on caution threshold is determined by the condition on caution setter 18. In the outputted information on caution, the information on caution outputter 19 includes information consistent with the condition on caution.
  • the alert system 1 includes a controller 20, a storage 30, and a communicator 40.
  • the controller 20 includes a CPU 21, a random access memory 22 (RAM), and a read only memory 23 (ROM).
  • the CPU 21 acts as each of the aforementioned operations units by readout of a program stored in the storage 30 to the RAM 22 and execution of the program.
  • the RAM 22 includes a volatile memory and is used as a working region by the CPU 21.
  • the ROM 23 includes a non-volatile memory such as an organic memory, and stores a control program executed by the CPU 21 for basic operations of this alert system 1, a basic input output system (BIOS), and the like.
  • the storage 30 includes a hard disc drive, flash memory, and the like, and contains various types of information and programs.
  • the storage 30 stores various types of information such as the samples extracted by the sample extractor 13, the prediction model constructed by the prediction model constructor 14, the alert output threshold, and the alert reason classification condition. Further, the storage 30 stores information on the prediction target disease, a standard for determination from the vital signs data whether the patients had the onset of a specific disease, and the like.
  • the communicator 40 communicably connects to the hospital network 5 and exchanges various types of information with the hospital network 5.
  • the alert system 1 executes the prediction model construction processing to construct the prediction model prior to the start of operation. Further, the prediction model construction processing is processing corresponding to machine learning prior to operation.
  • the vital signs data acquirer 11 acquires the vital signs data as illustrated in FIG. 6 (step S11). Further, the prediction model construction processing is started by an instruction such as a remote command execution instruction from a non-illustrated terminal used by the system manager.
  • the vital signs data acquired in step S11 is vital signs data of the patient accumulated in the hospital information system 3 prior to operation.
  • This step S11 is a step of acquisition of vital signs data to acquire the vital signs data of the patient.
  • the vital signs data acquirer 11 functions as a vital signs data acquisition means.
  • the onset information acquirer 12 acquires the onset information of the patient for which the vital signs data is previously acquired in step S11 (step S12). Specifically, on the basis of the determination standard stored in the storage 30, the onset information acquirer 12 determines whether, from the vital signs data, the respective patient had the onset of the prediction target disease. Further, as described above, the onset information acquirer 12 may acquire determinations by the physician of whether various patients had the onset of the disease.
  • This step S12 is a step of acquisition of onset information by acquiring onset information indicating whether the patient had an onset of the disease. Further, the onset information acquirer 12 in this step S12 functions as an onset information acquisition means.
  • the sample extractor 13 extracts the samples from the vital signs data acquired in step S11 (step S13).
  • the sample extractor 13 extracts each of the positive samples and the negative samples.
  • the sample extractor 13 extracts 50 each of the positive samples and the negative samples.
  • the sample extractor 13 extracts the positive samples and the negative samples by calculating the characteristic amount by applying the function as described above to the vital signs data to form the positive sample candidates and the negative sample candidates.
  • the sample extractor 13 stores in the storage 30 the positive samples and the negative samples extracted in this manner.
  • This step S13 is a sample extraction step for extraction of the samples.
  • the sample extractor 13 functions as a sample extraction means.
  • the prediction model constructor 14 constructs the prediction model on the basis of the positive samples and the negative samples extracted by the sample extractor 13 (step S14).
  • the prediction model constructor 14 selects the prediction model on the basis of the parameters for model selection stored in the storage 30, as described above.
  • the prediction model constructor 14 on the basis of the model parameters imparted to the prediction model, such as rules, equations, coefficients, and the like of the selected prediction model, calculates the prediction score for each sample extracted in step S13.
  • the prediction model constructor 14 stores the model parameters imparted to the prediction model, such as rules, equations, coefficients, and the like of the selected prediction model, and prediction score-associated samples in the storage 30.
  • This step S14 is a prediction model construction step for construction of the prediction model.
  • the prediction model constructor 14 functions as a prediction model construction means.
  • the prediction model constructor 14 determines the alert output threshold (step S15). In the case in which the central value of the prediction scores calculated in step S14 is 0.5, for example, the prediction model constructor 14 determines that the alert output threshold is 0.5, which is the central value of the prediction scores. Then the prediction model constructor 14 stores the determined alert output threshold in the storage 30.
  • the prediction model constructor 14 sets the alert reason classification condition (step S16).
  • the specific procedure for the prediction model constructor 14 to set the alert reason classification condition is described below.
  • the prediction model constructor 14 applies principal component analysis to the acquired positive samples and extracts the principal components. Specifically, in order of the principal component having the highest contribution rate contributing to the prediction of the onset of the disease, the prediction model constructor 14 extracts the principal components until a cumulative contribution rate of the extracted principal components reaches a predetermined cumulative contribution rate.
  • FIG. 7 illustrates an example of extraction of the principal components from the positive samples (record no. 1 to 3) illustrated in FIG. 4.
  • the contribution rate of a principal component 1 is 50%
  • the contribution rate of a principal component 2 is 20%
  • the contribution rate of a principal component 3 is 10%
  • the cumulative contribution rate (that is, the sum of the proportions of variance) of the three principal components is 80%, and thus these three principal components are extracted.
  • each principal component is weighted for each characteristic amount, and thus the degree of association between each of the characteristic amounts and the principal components is made clear.
  • the principal component 1 has the weightings of 0.8 for the average of blood pressure and 0.2 for the standard deviation of blood pressure.
  • the prediction model constructor 14 performs principal component analysis targeting each of the characteristic amounts of the positive samples. Thereafter, the prediction model constructor 14, as illustrated in FIG. 8, assigns scores for each of the principal components of the positive samples. Then the prediction model constructor 14, on the basis of each of the respective principal components, performs clustering and sets, as the alert reason classification condition, the condition imparting the classifications to each of the positive samples as illustrated in FIG. 9. Further, the imparted classifications may be set beforehand as parameters. Then the prediction model constructor 14 stores the set alert condition classification condition in the storage 30.
  • the alert system in this manner, constructs the prediction model and determines the alert output threshold.
  • the prediction model construction processing of setting the alert reason classification condition is executed in this manner.
  • the prediction model reconstruction processing performed during operation is described below with reference to FIG. 10.
  • the prediction model reconstruction processing reconstructs the prediction model by machine learning and updates the alert output threshold and the alert reason classification condition.
  • the alert system 1 executes the prediction model reconstruction processing periodically by night-time batch processing executed once per day, for example.
  • the vital signs data acquirer 11 acquires the vital signs data from the hospital information system 3 via the hospital network 5 (step S21).
  • This vital signs data is the vital signs data acquired by the medical instrument 2 during operation from the patient and accumulated in the hospital information system 3.
  • the onset information acquirer 12 acquires the onset information of the patients for which the vital signs data is acquired in step S21 (step S22). Specifically, the onset information acquirer 12, on the basis of the determination standards stored in the storage 30, determines from the vital signs data whether the respective patient had the onset of the prediction target disease. Further, the onset information acquirer 12 may acquire the onset information for each patient which the physician determines had the onset of the disease.
  • the action information acquirer 17 acquires the action information stored in the storage 30 with respect to the patient during operation (step S23).
  • the action information is selected information (selected as one choice from among "administer pharmaceutical", “continue monitoring”, and “no problem"), the patient ID, and the date-time of generation of the alert.
  • the action information acquirer 17 functions as an action information acquisition means for acquisition of action information indicating content of actions performed with respect to the prediction target patient for which the onset prediction means predicts the onset of the disease.
  • the sample extractor 13 extracts the samples from the vital signs data acquired in step S21 (step S24).
  • the sample extractor 13 extracts, for example, 25 each of the positive samples and the negative samples.
  • the sample extractor 13 updates 25 samples each of the positive samples to these samples extracted in step S24.
  • the sample extractor 13 stores in the storage 30 the positive samples and the negative samples having characteristic amounts calculated from the vital signs data as illustrated in FIG. 4.
  • the sample extractor 13 preferentially extracts portions of the samples by a below-described method on the basis of the action information acquired by the action information acquirer 17. Further, in the aforementioned example, the sample extractor 13, from among the samples extracted in step S11 of the prediction model construction processing, updates 50% in this step S24. Further, the proportion of updating is not necessarily 50%, and the proportion of updating may be stored beforehand as a parameter in the storage 30. In this case, the sample extractor 13 acquires the proportion of updating from the storage 30 and updates the samples using the acquired proportion. Further, rather than updating the samples, the sample extractor 13 may newly add samples.
  • the aforementioned preferential sample extraction method is described below concretely with reference to FIG. 11.
  • the "false negative sample” is a sample in the case of a patient, for which the onset of the disease is not previously predicted within the alert target time period, and for which the onset of the disease actually occurs in the alert target time period.
  • the sample extractor 13 extracts as a positive sample the vital signs data of such a patient from the time of generation of the alert to elapsing of the alert time period, and the sample extractor 13 sets the degree of priority to 1.
  • the term "degree of priority” means the degree of priority of extraction, with extraction being prioritized for the samples having the highest value of the degree of priority.
  • the extracted number is greater that of the samples of the degree of priority 1, for example, resulting in extraction so that the number of degree of priority 2 samples is twice the number of degree of priority 1 samples. Further, the degree of priority is set to 0 for the samples that are not set in FIG. 11.
  • the "false positive sample” is a sample in the case of a patient, for which the onset of the disease is previously predicted within the alert target time period, and for which the onset of the disease actually does not occur in the alert target time period. Due to prediction of the onset of the disease in this patient, the alert system 1 is expected to have output the alert. In such a case, the alert system 1, due to the action selected by the physician, changes the treatment and degree of priority of the extracted sample. Specifically, firstly the sample extractor 13 treats as a positive sample the sample for which "administer pharmaceutical" is selected, and the degree of priority is set to 2.
  • the sample extractor 13 treats the sample as a negative sample and sets the degree of priority to 1. Then for the samples for which "no problem” is selected, the sample extractor 13 treats the sample as a negative sample and sets the degree of priority to 2.
  • the sample extractor 13 extracts the samples in accordance with the degrees of priority set in this manner. Further, rather than setting the aforementioned degree of priority, a fraction of extraction as samples may be set. However, as previously described, a balance is preferably maintained between the sample numbers of the positive samples and the negative samples. Thus in a range capable of maintaining this balance, the sample extractor 13 actually extracts samples as set by the degree of priority or fraction.
  • the prediction model constructor 14 on the basis of the acquired information, constructs the prediction model and reconstructs the prediction model stored in the storage 30 (step S25).
  • the construction technique of the prediction model is similar to that of step S14 of the prediction model construction processing.
  • the prediction model constructor 14 reconstructs the prediction model by replacement with the reconstructed prediction model. Further, the prediction model constructor 14 may perform a comparison between the constructed prediction model and the prediction model stored in the storage 30, and may update the prediction model by updating the differences.
  • the prediction model constructor 14 updates the alert output threshold (step S26). Specifically, in the case in which there are many false positive samples, in order to make the output condition for the alert more severe to lower the number of outputted alerts, the prediction model constructor 14 increases the alert output threshold. Conversely, in the case in which there are many false negative samples, the prediction model constructor 14 decreases the alert output threshold.
  • the prediction model constructor 14 updates the alert reason classification condition (step S27).
  • the basic processing procedure is similar to that of the step S16 of the prediction model construction processing illustrated in FIG. 6.
  • the prediction model constructor 14 sets the alert reason classification condition and updates the alert reason classification condition. Further, this updating may be performed by replacement with the set alert reason classification condition. Further, the prediction model constructor 14 may perform a comparison between the set alert reason classification condition and the alert reason classification condition stored in the storage 30, and may update the alert reason classification condition by updating the difference.
  • the condition on caution setter 18 sets or updates the condition on caution (step S28).
  • the condition on caution setter 18 extracts the vital signs data acquired by the vital signs data acquirer 11 in step S21 and for which the action information acquired by the action information acquirer 17 in step S23 is "no problem". Then the condition on caution setter 18 calculates the characteristic amounts on the basis of the extracted vital signs data. Then the condition on caution setter 18 sets the condition on caution by machine learning, such as by decision tree analysis, using the characteristic amounts of patients having the onset of the disease within the standard time period and the characteristic amounts of the patients having no onset of the disease. In this case, the conditions may be set by a rule base rather than by machine learning.
  • the condition on caution setter 18 sets the condition on caution to indicate a condition for calculation of the score on caution, such as by setting the score on caution to 0.9 when the average blood pressure is greater than or equal to 140 mmHg, for example. Then the condition on caution setter 18 stores the set condition on caution in the storage 30. Further, in the case in which the condition on caution is previously stored in the storage 30, the condition on caution setter 18 updates the condition on caution by replacement with the set condition on caution. Further, the condition on caution setter 18 may perform a comparison between the set condition on caution and the condition on caution stored in the storage 30, and may update the condition on caution by updating the difference.
  • the condition on caution setter 18 determines the information on caution threshold (step S29).
  • the condition on caution setter 18 determines the information on caution threshold from a previously set parameter. Further, the condition on caution setter 18 may calculate the scores on caution concerning the characteristic amounts of the patient calculated in step S28, and may set the threshold to the central value of the calculated scores on caution. Then the condition on caution setter 18 stores the determined information on caution threshold in the storage 30.
  • the alert system 1 in the aforementioned manner reconstructs the prediction model and updates the alert output threshold and the alert reason classification condition. Then the alert system 1 sets or updates the condition on caution and determines the information on caution threshold.
  • the alert system 1 in this manner can raise the accuracy of prediction of the onset of the disease during operation.
  • the alert processing for prediction of disease of the patient by the alert system 1 to output the alert is described below with reference to FIG. 12.
  • the alert system 1 executes this alert processing to specify the target patient and the target disease periodically, such as every 5 minutes, for example.
  • the time interval between such executions is preferably about that of the aforementioned sampling interval.
  • the vital signs data acquirer 11 acquires the vital signs data (step S301).
  • the vital signs data is acquired by receiving from the medical instrument 2 the vital signs data transmitted in real time.
  • the vital signs data acquirer 11 stores the received vital signs data in the storage 30 and retains the received vital signs data for at least 30 minutes, that is, the characteristic amount window time. Then the vital signs data acquirer 11 combines the retained vital signs data and the received vital signs data, and calculates the characteristic amounts for each patient on the basis of a 30 minute portion of the vital signs data.
  • the onset predictor 15 applies the prediction model to the characteristic amounts calculated by the vital signs data acquirer 11 in step S301, and calculates the prediction score indicating the probability of the occurrence of the disease within the alert target time period (step S302).
  • This prediction model is the prediction model constructed in the prediction model construction processing of step S14, or alternatively, is the reconstructed prediction model reconstructed in the prediction model reconstruction processing of step S25. Further, the processing from step S302 to a below-described step S310 is executed patient by patient.
  • the onset predictor 15 determines whether the prediction score calculated by the onset predictor 15 in step S302 is greater than or equal to the alert output threshold (step S303). Further, this alert output threshold is the alert output threshold determined in step S15 of the prediction model construction processing, or alternatively, is the alert output threshold determined in step S26 of the prediction model reconstruction processing.
  • This step S303 is an onset prediction step for prediction of whether the prediction target patient will have the onset of the disease within the standard time period. In this step S303, the onset predictor 15 functions as an onset prediction means.
  • the alert information outputter 16 When the onset predictor 15 determines that the prediction score is greater than or equal to the threshold (YES in step S303), the alert information outputter 16 generates the alert information (step S304). Specifically, the alert information outputter 16, on the basis of the alert reason classification condition, calculates the principal component scores of the vital signs data, and applies classifications. Further, this alert reason classification condition is constructed in step S16 of the prediction model construction processing, or alternatively, is the alert reason classification condition updated in step S27 of the prediction model reconstruction processing. Then the alert information outputter 16 prepares the alert information that includes the calculated principal component scores and the imparted classifications. Further, the alert information may include the characteristic amounts or the vital signs data related to the calculated principal components.
  • the controller 20 of the alert system 1 determines whether processing is previously performed concerning all the prediction target patients (step S311). When the controller 20 determines that the processing is performed for all the prediction target patients (YES in step S311), the controller 20 ends the alert processing. However, when the controller 20 determines that the processing is not performed for all the prediction target patients (NO in step S311), the onset predictor 15 returns to the processing of step S302 to calculate the prediction score of the next prediction target patient. Further, although the present example indicates in-order processing of the data of the patients one patient at a time, the controller 20 may process the data of multiple patients in parallel by distributed execution of multiple processes.
  • the alert information outputter 16 transmits the generated alert information via the hospital network 5 to the portable terminal 4 (step S305).
  • the portable terminal 4 in addition to generating a notifying sound and notifying the physician of the receiving of the alert information, displays on a screen the received alert information. As illustrated in FIG. 13, the screen displayed by the portable terminal 4 visually displays information such as the target disease, the prediction score, the principal component score, and the imparted classification.
  • the portable terminal 4 on the basis of the received alert information, generates this type of screen. Further, the portable terminal 4 may display the received alert information by the alert information outputter generating and transmitting the alert information for display on the screen.
  • the processing to convert to information for visual display the information such as the target disease, the prediction score, the principal component score, and the imparted classification included in the alert information may be performed by the alert system 1 or may be performed by the portable terminal 4. Further, a portion of the conversion processing may be performed by the alert system 1, and the other conversion processing may be performed by the portable terminal 4. Further, the portable terminal 4 may display on the screen information such as the vital signs data or the characteristic amounts relating to the principal components.
  • the alert information outputter 16 functions as an alert information output means for outputting alert information imparting classifications by applying principal component analysis to characteristic amounts expressing characteristics of the vital signs data or the vital signs data of the patient which is the prediction target predicted by the onset prediction means to have the onset of the disease.
  • the portable terminal 4 displays choices of actions such as “administer pharmaceutical”, “continue monitoring”, and “no problem” as illustrated in FIG. 14.
  • the physician understands the status of the patient on the basis of the information displayed on the portable terminal 4 and determines the action to be taken. Then the physician operates the portable terminal 4 and selects a displayed choice.
  • the portable terminal 4 transmits the action information indicating the selected action via the hospital network 5 to the alert system 1 and the hospital information system 3.
  • the action information acquirer 17 acquires the action information by receiving the action information via the hospital network 5 (step S306).
  • the action information is transmitted from the portable terminal 4 in real time.
  • This action information is the action information of the prediction target, referred to hereinafter as the "prediction target action information".
  • the action information acquirer 17 associates the prediction target action information with the target patient ID and the day and time of generation of the alert, and then stores the resulting information in the storage 30.
  • the information on caution outputter 19 determines whether the prediction target action information acquired by the action information acquirer 17 is "no problem" (step S307). Upon determination that the action information is "no problem" (YES in step S307), the information on caution outputter 19 calculates the score on caution (step S308). Specifically, in accordance with the condition on caution set or updated in step S28 of the prediction model reconstruction processing, the information on caution outputter 19 analyses the characteristic amounts of the patient calculated by the vital signs data acquirer 11 in step S301 and calculates the score on caution.
  • the score on caution of the patient for which the average blood pressure is 143 mmHg is calculated to be 0.9 due to conformance to this condition on caution.
  • step S307 the controller 20 of the alert system 1 proceeds to the processing of step S311 that determines whether processing is previously performed for all the prediction target patients.
  • the information on caution outputter 19 determines whether the score on caution is greater than or equal to the information on caution threshold (step S309). Further, the information on caution threshold is determined in step S29 of the prediction model reconstruction processing. For example, in the case in which the information on caution threshold is 0.5 and the calculated score on caution is 0.9, the information on caution outputter 19 determines that the score on caution is greater than or equal to the information on caution threshold. Upon determination that the score on caution is greater than or equal to the information on caution threshold (YES in step S309), the information on caution outputter 19 generates the information on caution and transmits the information on caution via the hospital network 5 to the portable terminal 4 (step S310). The processing returns to step S301 in which the vital signs data acquirer 11 acquires the vital signs data.
  • step S309 the controller 20 of the alert system 1 proceeds to the processing of step S311 in which determination is made as to whether there is prior processing for all the prediction target patients.
  • step S310 the information on caution outputter 19 functions as an information on caution output means that outputs the information on caution.
  • the information on caution includes information such as the matched condition on caution and the calculated score on caution. Also, the portable terminal 4 displays the received information on caution. Further, in the same manner as the alert information, the processing to convert the information on caution into screen display information may be performed by the alert system 1 or by the portable terminal 4. Further, a portion of the conversion processing may be performed by the alert system 1, and the other conversion processing may be performed by the portable terminal 4.
  • the alert system 1 applies the prediction model and predicts whether the target patient will have the onset of the specific disease. Then in the case in which the onset of the disease is predicted, the alert system 1 transmits the alert information to the portable terminal 4 via the hospital network 5. Further, the alert system 1 acquires the action information inputted to the portable terminal 4, and in the case in which the action is "no problem", determines whether to transmit the information on caution. In the case in which the determination is to transmit the information on caution, the alert system 1 transmits the information on caution to the portable terminal 4 via the hospital network 5.
  • the alert system 1 of the present embodiment can predict the onset of the disease up to several hours prior to the onset and can transmit the prediction to the physician.
  • signs of the disease onset in fact often appear several hours prior to the disease onset.
  • the sample extractor 13 of the alert system by the setting of the alert target time period to greater than or equal to several hours, extracts as the vital signs data to form the positive samples the blood pressures of the characteristic amount window time period illustrated in FIG. 15A.
  • the alert system 1 calculates as the characteristic amounts the slopes of the blood pressure in this range, uses these characteristic amounts as the positive samples for learning, and constructs the prediction model on the basis of the slopes of the blood pressure.
  • Such construction enables prediction of the onset of disease for a target patient in which signs appear that are similar to those of the patient A.
  • the variance or standard deviation of the range of the characteristic amount window time period is used as the characteristic amount for learning to construct the prediction model on the basis of the variance or standard deviation of blood pressure.
  • Such construction enables prediction of the onset of disease for a target patient in which signs appear that are similar to those of the patient B.
  • the alert system 1 of the present embodiment among the vital signs data of the patient had an onset of the disease, extracts the positive samples from the vital signs data measured from a standard time period prior to an onset time of the disease until the onset time of the disease, extracts the negative samples from the vital signs data measured earlier than the standard time period before the onset time, and uses the extracted samples for learning. Due to such learning, the alert system 1 can learn by comparison between the characteristics appearing in the vital signs data measured from a standard time period prior to an onset time of the disease until the onset time of the disease and the characteristics appearing in the vital signs data measured earlier than the standard time period before the onset time, and can find signs of the occurrence of the disease onset.
  • the alert system 1 extracts the vital signs data measured during a total time period, which is a sum of the alert target time period of the onset time and a target-exclusion time period, starting from the onset time. Then for the vital signs data measured in this time period, the alert system 1 for each characteristic amount window time period performs multiple extractions of samples forming the positive samples, such that a portion of these characteristic amount window time periods overlap. Such overlapping enables finding of the signs of disease onset without overlooking of signs, and enables accurate prediction of the onset of the disease. Further, in order to avoid overlapping acquisition of the vital signs data forming the positive sample and the vital signs data forming the negative sample, the aforementioned target-exclusion time period is preferably set to the same length as that of the characteristic amount window time period.
  • the action information selected by the physician is reflected in the method of extraction used by the sample extractor 13 of the alert system 1 of the present embodiment. Due to such configuration, the alert system 1 can make use of the content of the response to the patient for improvement of the accuracy of machine learning. In particular, even in the case in which the extracted sample is a false positive sample in which the patient which is predicted to have the disease onset had no onset of the disease within the alert target time period, the alert system 1 treats the extracted sample as a positive sample rather than a negative sample when the physician receiving an alert selects the "administer pharmaceutical" action.
  • the alert system 1 extracts the false positive sample in a preferential manner as a negative sample.
  • the alert system 1 can prioritize learning to determine corrections to be made concerning the vital signs data for which the prior determination is erroneous.
  • the alert system 1 of the present embodiment uses the alert reason classification conditions to classify the component indicating the reason of the disease onset prediction, and transmits the component as the alert information to the portable terminal 4 via the hospital network 5.
  • the physician can obtain information serving as an indicator for investigating the response to the target patient.
  • the alert system 1 of the present embodiment In the case in which the action for the patient for which the alert information is generated is "no problem", the alert system 1 of the present embodiment generates the information on caution and transmits the information on caution to the portable terminal 4 via the hospital network 5. Such operation gives the physician an opportunity to notice an erroneous determination.
  • the vital signs data of the present embodiment may also include information on pharmaceuticals administered to the patient, treatments, environmental information, and the like.
  • the prediction of the disease onset, the generation of the alert information, the generation of the information on caution, and the like can be performed with increased accuracy.
  • the physician viewing the alert information can understand what type of pharmaceuticals may be administered.
  • samples may be used that are neither positive samples nor negative samples.
  • the samples used for learning may be samples that have indeterminate positivity-negativity, that is, samples for which positivity and negativity cannot be determined. Due to enablement by this means of learning concerning the characteristic amount for indeterminate positivity-negativity, a prediction model can be constructed that includes ambiguity.
  • the alert system 1 predicts whether the prediction target patient will have the onset of the disease by calculating the prediction score.
  • the scope of the present disclosure is not limited to this configuration, and as long as whether the onset of the disease will occur in the prediction target patient is predicted, the alert system 1 may predict whether the disease will be onset in the prediction target patent by direct calculation rather than by calculating the prediction score. That is to say, the alarm system 1 may calculate, for example, a prediction value indicating only whether the target patient will have the onset of the disease, such as 1, which indicates that the prediction target patient will have the onset of the disease, or 0, which indicates that the prediction target patient will not have the onset of the disease.
  • the intercommunication between the alert system 1, the medical instrument 2, the hospital information system 3, and the portable terminal 4 is performed via the hospital network 5.
  • the scope of the present disclosure is not limited to this configuration, and intercommunication may be performed directly through dedicated wiring and the like.
  • the alert system 1 may be contained in or integrated with the medical instrument 2 or the hospital information system 3.
  • the present embodiment provides an example of operation of the alert system 1 within a hospital.
  • the present disclosure is not limited to this configuration, and for example, a web site may be utilized, and if the necessary vital signs data can be obtained using a wearable sensor, the alert system 1 may be operated outside of the hospital, such as in a home.
  • the alert system 1 can be accomplished by using a normal computer and without depending on a dedicated device.
  • an alert system 1 executing the aforementioned processing may be configured by installing on a computer a program, for execution of the above processing by the computer, that is stored on a recording medium.
  • a single alert system 1 may be configured by cooperative operation of multiple computers.
  • any desired method may be used for providing the program to the computer.
  • the program may be provided via a communication line, a communication network, a communication system, and the like.
  • OS operating system
  • Alert system 2 Medical instrument 3 Hospital information system 4 Portable terminal 5 Hospital network 11 Vital signs data acquirer 12 Onset information acquirer 13 Sample extractor 14 Prediction model constructor 15 Onset predictor 16 Alert information outputter 17 Action information acquirer 18 Condition on caution setter 19 Information on caution outputter 20 Controller 21 CPU 22 RAM 23 ROM 30 Storage 40 Communicator 101, 102, 103, 104, 105, 106, 107 Double-headed arrow indicating target period of positive samples 108, 109, 110, 111 Double-headed arrow indicating target period of negative samples

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pathology (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Surgery (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Veterinary Medicine (AREA)
  • Animal Behavior & Ethology (AREA)
  • Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physiology (AREA)
  • Signal Processing (AREA)
  • Psychiatry (AREA)
  • Business, Economics & Management (AREA)
  • General Business, Economics & Management (AREA)
  • Immunology (AREA)
  • Vascular Medicine (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

Un système d'alerte (1) comprend un extracteur d'échantillon (13) et un constructeur de modèle de prédiction (14). L'extracteur d'échantillon (13) extrait, lorsque les informations d'apparition indiquent que la maladie est apparue chez le patient, (i) des données de signes vitaux pour obtenir des échantillons positifs à partir de et parmi des données de signes vitaux de construction de modèle de prédiction, les données de signes vitaux étant mesurées à partir d'une période de temps standard avant le moment d'apparition de la maladie jusqu'au moment d'apparition de la maladie, et (ii) des données de signes vitaux pour obtenir des échantillons négatifs à partir des données de signes vitaux, les données de signes vitaux étant mesurées antérieurement à la période de temps standard avant le moment d'apparition.
PCT/JP2018/000023 2017-07-07 2018-01-04 Dispositif de prédiction d'apparition de maladie, procédé de prédiction d'apparition de maladie et programme WO2019008798A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2017-134050 2017-07-07
JP2017134050A JP6909078B2 (ja) 2017-07-07 2017-07-07 疾病発症予測装置、疾病発症予測方法およびプログラム

Publications (1)

Publication Number Publication Date
WO2019008798A1 true WO2019008798A1 (fr) 2019-01-10

Family

ID=61028136

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2018/000023 WO2019008798A1 (fr) 2017-07-07 2018-01-04 Dispositif de prédiction d'apparition de maladie, procédé de prédiction d'apparition de maladie et programme

Country Status (2)

Country Link
JP (1) JP6909078B2 (fr)
WO (1) WO2019008798A1 (fr)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110974215A (zh) * 2019-12-20 2020-04-10 首都医科大学宣武医院 基于无线心电监护传感器组的预警系统及方法
CN110974216A (zh) * 2019-12-20 2020-04-10 首都医科大学宣武医院 一种无线心电监护传感器的遥控系统
CN110993103A (zh) * 2019-11-28 2020-04-10 阳光人寿保险股份有限公司 疾病风险预测模型的建立方法和疾病保险产品的推荐方法
US20210193317A1 (en) * 2019-12-20 2021-06-24 Fresenius Medical Care Holdings, Inc. Real-time intradialytic hypotension prediction
CN113017572A (zh) * 2021-03-17 2021-06-25 上海交通大学医学院附属瑞金医院 一种重症预警方法、装置、电子设备及存储介质
CN113057587A (zh) * 2021-03-17 2021-07-02 上海电气集团股份有限公司 一种病症预警方法、装置、电子设备及存储介质
CN113782182A (zh) * 2021-07-29 2021-12-10 北京理工大学 一种基于肌电数据的应力性骨折预测方法
CN114159070A (zh) * 2021-12-20 2022-03-11 武汉大学 一种卷积神经网络的心脏骤停风险实时预测方法及系统
US20220084635A1 (en) * 2020-09-15 2022-03-17 Acer Incorporated Disease classification method and disease classification device
CN114343575A (zh) * 2021-12-27 2022-04-15 福寿康(上海)医疗养老服务有限公司 一种基于家床设备报警信息的统计分析方法
CN115719647A (zh) * 2023-01-09 2023-02-28 之江实验室 融合主动学习和对比学习的血透并发心血管疾病预测系统

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7238910B2 (ja) * 2019-02-08 2023-03-14 日本電気株式会社 生体情報処理装置、方法、及びプログラム
JP7027359B2 (ja) * 2019-02-20 2022-03-01 株式会社日立製作所 ヘルスケアデータ分析装置及びヘルスケアデータ分析方法
CN110610767B (zh) * 2019-08-01 2023-06-02 平安科技（深圳）有限公司 发病率监测方法、装置、设备及存储介质
JP7399456B2 (ja) * 2019-11-07 2023-12-18 国立研究開発法人物質・材料研究機構 システム、サーバ装置、クライアント装置、バイオセンサ、バイオセンサセット、データ取得装置、及び、プログラム
JP7427452B2 (ja) * 2020-01-14 2024-02-05 エヌ・ティ・ティ・コミュニケーションズ株式会社 情報処理装置、発症確率判定方法および発症確率判定プログラム
JP2021176424A (ja) * 2020-05-08 2021-11-11 コニカミノルタ株式会社 医療診断支援装置、医療診断支援プログラム、および医療診断支援方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030181815A1 (en) * 2002-03-22 2003-09-25 Ebner Dennis M. Method for continuous monitoring of patients to detect the potential onset of sepsis
US20080103996A1 (en) * 2006-10-31 2008-05-01 George Forman Retraining a machine-learning classifier using re-labeled training samples
JP2009517188A (ja) 2005-11-29 2009-04-30 ベンチャー ゲイン リミテッド ライアビリティー カンパニー ヒトの健康に関する残差ベースの管理
US20090163776A1 (en) * 2005-07-18 2009-06-25 Integralis Ltd. Apparatus, method and computer readable code for forecasting the onset of potentially life-threatening disease
JP2016040693A (ja) 2014-08-13 2016-03-24 日本光電工業株式会社 疾病予測装置、疾病予測方法、及びプログラム

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002143097A (ja) * 2000-11-10 2002-05-21 Kuniaki Otsuka 疾病発症予知通報システム及び方法
JP2006285425A (ja) * 2005-03-31 2006-10-19 Mitsubishi Heavy Ind Ltd ロボットを用いたヘルスケアサービスシステム

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030181815A1 (en) * 2002-03-22 2003-09-25 Ebner Dennis M. Method for continuous monitoring of patients to detect the potential onset of sepsis
US20090163776A1 (en) * 2005-07-18 2009-06-25 Integralis Ltd. Apparatus, method and computer readable code for forecasting the onset of potentially life-threatening disease
JP2009517188A (ja) 2005-11-29 2009-04-30 ベンチャー ゲイン リミテッド ライアビリティー カンパニー ヒトの健康に関する残差ベースの管理
US20080103996A1 (en) * 2006-10-31 2008-05-01 George Forman Retraining a machine-learning classifier using re-labeled training samples
JP2016040693A (ja) 2014-08-13 2016-03-24 日本光電工業株式会社 疾病予測装置、疾病予測方法、及びプログラム

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
AHLAME DOUZAL-CHOUAKRIA ET AL: "Classification trees for time series", PATTERN RECOGNITION, ELSEVIER, GB, vol. 45, no. 3, 12 August 2011 (2011-08-12), pages 1076 - 1091, XP028326553, ISSN: 0031-3203, [retrieved on 20110823], DOI: 10.1016/J.PATCOG.2011.08.018 *
CARLOS J ALONSO GONZÁLEZ ET AL: "Boosting Interval-Based Literals: Variable Length and Early Classification", 29 July 2015 (2015-07-29), XP055462464, Retrieved from the Internet <URL:https://pdfs.semanticscholar.org/9481/52b25cfac9acc8226d0da54b97711bf80049.pdf> [retrieved on 20180323] *
DANIIL RYABKO ET AL: "A binary-classification-based metric between time-series distributions and its use in statistical and learning problems", JOURNAL OF MACHINE LEARNING RESEARCH, MIT PRESS, CAMBRIDGE, MA, US, vol. 14, no. 1, 30 September 2013 (2013-09-30), pages 2837 - 2856, XP058035596, ISSN: 1532-4435 *

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110993103A (zh) * 2019-11-28 2020-04-10 阳光人寿保险股份有限公司 疾病风险预测模型的建立方法和疾病保险产品的推荐方法
CN110974216A (zh) * 2019-12-20 2020-04-10 首都医科大学宣武医院 一种无线心电监护传感器的遥控系统
US20210193317A1 (en) * 2019-12-20 2021-06-24 Fresenius Medical Care Holdings, Inc. Real-time intradialytic hypotension prediction
CN110974215A (zh) * 2019-12-20 2020-04-10 首都医科大学宣武医院 基于无线心电监护传感器组的预警系统及方法
US11830589B2 (en) * 2020-09-15 2023-11-28 Acer Incorporated Disease classification method and disease classification device
US20220084635A1 (en) * 2020-09-15 2022-03-17 Acer Incorporated Disease classification method and disease classification device
CN113017572A (zh) * 2021-03-17 2021-06-25 上海交通大学医学院附属瑞金医院 一种重症预警方法、装置、电子设备及存储介质
CN113057587A (zh) * 2021-03-17 2021-07-02 上海电气集团股份有限公司 一种病症预警方法、装置、电子设备及存储介质
CN113017572B (zh) * 2021-03-17 2023-11-28 上海交通大学医学院附属瑞金医院 一种重症预警方法、装置、电子设备及存储介质
CN113782182A (zh) * 2021-07-29 2021-12-10 北京理工大学 一种基于肌电数据的应力性骨折预测方法
CN114159070A (zh) * 2021-12-20 2022-03-11 武汉大学 一种卷积神经网络的心脏骤停风险实时预测方法及系统
CN114343575A (zh) * 2021-12-27 2022-04-15 福寿康(上海)医疗养老服务有限公司 一种基于家床设备报警信息的统计分析方法
CN115719647A (zh) * 2023-01-09 2023-02-28 之江实验室 融合主动学习和对比学习的血透并发心血管疾病预测系统

Also Published As

Publication number Publication date
JP6909078B2 (ja) 2021-07-28
JP2019016235A (ja) 2019-01-31

Similar Documents

Publication Publication Date Title
WO2019008798A1 (fr) Dispositif de prédiction d&#39;apparition de maladie, procédé de prédiction d&#39;apparition de maladie et programme
Edwards et al. Development and validation of a risk prediction model for in-hospital mortality after transcatheter aortic valve replacement
US11881293B2 (en) Methods for automatic cohort selection in epidemiologic studies and clinical trials
WO2019103908A1 (fr) Collecte automatisée d&#39;informations et traitement de données cliniques
KR20180064952A (ko) 빅데이터를 활용한 건강정보 예측 장치 및 방법
JP5977898B1 (ja) 行動予測装置、行動予測装置の制御方法、および行動予測装置の制御プログラム
US11923094B2 (en) Monitoring predictive models
US20190311809A1 (en) Patient status monitor and method of monitoring patient status
JP2008546117A (ja) 疾患予後の動的定量のためのシステムおよび方法
KR20190132290A (ko) 환자 진단 학습 방법, 서버 및 프로그램
JP2020537232A (ja) 全母集団から任意に選択された部分母集団内の被験者における非健康状態のリスク、発生又は進行を予測する医療装置及びコンピュータ実装方法
WO2018106481A1 (fr) Procédés, systèmes implémentés par ordinateur et supports informatiques pour diagnostiquer un état de santé
Houlton How artificial intelligence is transforming healthcare
US20230245779A1 (en) System and method for peri-anaesthetic risk evaluation
KR101744800B1 (ko) 의료 정보 제공 시스템
CN113470825A (zh) 一种静脉血栓栓塞预测系统、方法、设备及介质
CN113808724B (zh) 数据分析方法及装置、存储介质及电子终端
JP2020013246A (ja) 疾病予測装置、方法およびプログラム
US20240120107A1 (en) Data processing system and method for predicting a score representative of a probability of a sepsis for a patient
KR102511516B1 (ko) 인공지능을 이용한 ct 조영제 유발 신독성 예측 장치 및 그 방법
US20230207125A1 (en) Diagnosis-adaptive patient acuity monitoring
US20200321128A1 (en) Data processing device, data processing method and non-transitory storage medium storing data processing program
KR20160123150A (ko) 의료 정보 제공 시스템
WO2022170215A1 (fr) Outil de visualisation d&#39;aperçu contextuel clinique et d&#39;assistance à la décision
Eshelman et al. A Methodology for Evaluating the Performance of Alerting and Detection Algorithms Running on Continuous Patient Data

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18701604

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18701604

Country of ref document: EP

Kind code of ref document: A1