JP7466865B2 - Self-removal occurrence prediction device, self-removal occurrence prediction method, program, fall occurrence prediction device, fall occurrence prediction method, and medical safety improvement support method - Google Patents

Description

The present invention relates to a self-removal prediction device, a self-removal prediction method, a program, a fall prediction device, a fall prediction method, and a medical safety improvement support method.

One of the most common medical accidents that occurs within hospitals and can pose serious problems for the physical and mental health of patients and the workload of medical staff is the removal of intravenous drips or drains by patients themselves.
Although there is some knowledge about factors related to self-removal of intravenous drips and drains in intensive care units, there is no technology that would enable this in general wards.

There are many medical accidents occurring in hospitals, such as elderly patients and patients with dementia or delirium removing drains or intravenous drips themselves, or falling or tripping, but there is no highly accurate method for predicting such medical accidents that meets the needs of the medical field.

As we enter a super-aging society, there are many medical accidents occurring in hospitals, such as patients self-removing drains and IV drips, or falling, due to frailty, dementia, delirium, etc. These medical accidents not only contribute to the worsening of the underlying disease, prolonging hospital stays and preventing patients from being transferred to home care, but can also sometimes lead to serious outcomes such as being bedridden or in a life-threatening state. Furthermore, when such medical accidents occur, doctors, nurses, and other medical professionals must expend a lot of effort evaluating the patient's condition after the accident, administering medical treatment, and consulting with specialists about the condition, placing a serious burden on hospital staff.
Given the above background, checklists and other tools have been developed to predict medical accidents that occur in hospitals, but their accuracy is insufficient and they are still a long way from being implemented in society.
The biggest factor in the occurrence of such problems is the sharp increase in the number of elderly people hospitalized with frailty, dementia, and delirium, which can cause medical accidents. In addition, the advancement of medical care has led to the development of various drugs and medical devices, and their use in the medical field is also a factor.

Patent Document 1 describes a data analysis system that predicts not only patient falls, but also the occurrence of accidents and dangerous behaviors due to the patient's actions or circumstances, such as falling off a bed. Patent Document 1 also describes that the data analysis system obtains information that contributes to regression from electronic medical records as a score through predictive coding processing for logistic regression analysis.

International Publication No. 2017/199445

Through intensive research, the inventors have discovered that when predicting the occurrence of a patient's fall using a predictive model algorithm (an algorithm that does not suppress overlearning) such as the technology described in Patent Document 1, overlearning occurs during the learning stage of the predictive model, and it is not possible to predict the occurrence of a patient's fall with high accuracy.
In addition, although the technology described in Patent Document 1 can predict the occurrence of a patient's fall, it cannot accurately predict the occurrence of a patient's self-removal of a drain or other device.

In consideration of the above-mentioned problems, the present invention has an object to provide an apparatus, a method, and a program for predicting the occurrence of self-removal of lines, which are the self-removal of at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted inside the body by a hospitalized patient, and which are capable of predicting with high accuracy whether or not such an occurrence will occur.
In addition, the present invention aims to provide a fall/slip prediction device, a fall/slip prediction method, and a program that can predict with high accuracy whether or not a fall will occur, which is when a patient slips or falls, within a specified period of time during their hospitalization.
In other words, an object of the present invention is to provide a medical safety improvement support method capable of supporting the improvement of medical safety.

One aspect of the present invention is a self-removal prediction device for predicting whether or not a hospitalized patient will self-remove a line, which is at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body, and the device comprises a prediction unit for predicting whether or not a line will self-removal will occur, and a learning unit for performing supervised learning on the prediction unit. The supervised learning performed by the learning unit includes supervised data including electronic medical record data of a self-removal occurrence group, which is a group including hospitalized patients who have been recorded as having self-removed or accidentally removed in the past, and control data including electronic medical record data of a self-removal occurrence group, which is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the self-removal occurrence group, but who are not included in the self-removal occurrence group. the electronic medical record data of the self-removal occurrence group includes attribute data including at least the gender, age, and medical department of the patients included in the self-removal occurrence group, and the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group; the prediction unit, in which supervised learning has been performed by the learning unit, predicts whether the prediction target patient will self-remove lines within a predetermined period of hospitalization based on the electronic medical record data of the prediction target patient who is hospitalized, and the electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient.

In one embodiment of the self-removal prediction device of the present invention, the electronic medical record data of the self-removal group includes attribute data including at least the gender, age, and medical department of the patients in the self-removal group, and drug-specific usage data indicating the usage of drugs by the patients in the self-removal group, the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients in the control group, and drug-specific usage data indicating the usage of drugs by the patients in the control group, and the electronic medical record data of the prediction target patient may include attribute data including at least the gender, age, and medical department of the prediction target patient, and drug-specific usage data indicating the usage of drugs by the prediction target patient.

In one embodiment of the self-removal prediction device of the present invention, L1 regularized logistic regression may be used as the algorithm of the self-removal prediction device.

In one embodiment of the self-removal prediction device of the present invention, the prediction unit may weight the prediction of whether the patient to be predicted will self-remove lines within the specified period differently depending on whether the drug included in the drug-specific usage data indicating the drug usage by the patient to be predicted is a psychotropic drug.

In one aspect of the self-removal prediction device of the present invention, the prediction unit may predict that the patient to be predicted will self-remove lines within the specified period when the drug-specific usage data indicating the drug usage by the patient to be predicted includes a history of psychotropic drug usage by the patient to be predicted, compared to when the drug-specific usage data indicating the drug usage by the patient to be predicted does not include a history of psychotropic drug usage by the patient to be predicted.

In one embodiment of the self-removal prediction device of the present invention, the prediction unit may predict that the patient to be predicted is more likely to self-remove lines within the specified period when the patient to be predicted is male than when the patient to be predicted is female.

In one embodiment of the self-removal prediction device of the present invention, the prediction unit may predict that the patient is more likely to self-remove lines within the specified period when the patient is 65 years old or older than when the patient is younger than 65 years old.

One aspect of the present invention is a method for predicting whether or not a patient will self-removal of a line, which is the self-removal of at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body, by a hospitalized patient. The method includes a prediction step for predicting whether or not a patient will self-removal of a line, and a learning step for performing supervised learning before the prediction step is performed. The supervised learning used in the learning step includes electronic medical record data of a self-removal group, which is a group including hospitalized patients who have been recorded as having self-removed or accidentally removed in the past, and a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the self-removal group, but are not included in the self-removal group. and electronic medical record data of a control group in which self-removal has occurred, the electronic medical record data of the self-removal occurrence group includes attribute data including at least the gender, age, and medical department of the patients in the self-removal occurrence group, and the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients in the control group, and the prediction step, which is executed after the learning step, predicts whether the prediction target patient will self-remove lines within a predetermined period of hospitalization based on the electronic medical record data of the prediction target patient who is hospitalized, and the electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient.

One aspect of the present invention is a program for causing a computer to execute a prediction step of predicting whether or not a self-removal of a line, which is the self-removal of at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body, will occur, and a learning step of performing supervised learning before the prediction step is executed. The supervised learning used in the learning step includes supervised data of a self-removal occurrence group, which is a group including hospitalized patients who have been recorded in the past as having had a self-removal or accidental removal, and computer data of a group including hospitalized patients who are at least the same gender, age, and hospital ward as the patients included in the self-removal occurrence group, but are not included in the self-removal occurrence group. the electronic medical record data of the self-removal occurrence group includes attribute data including at least the gender, age, and medical department of the patients included in the self-removal occurrence group, and the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group; the prediction step, which is executed after the learning step is executed, predicts whether the prediction target patient will self-remove lines within a specified period of hospitalization based on the electronic medical record data of the prediction target patient who is hospitalized, and the electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient.

Through extensive research, the inventors have discovered that by using logistic regression as an algorithm for a predictive model that predicts whether or not a patient will fall during a specified period of hospitalization, overfitting can be suppressed during the learning stage of the predictive model.

One aspect of the present invention is a fall occurrence prediction device that predicts whether or not a fall will occur, which is when a patient falls or falls down a certain period of time during hospitalization, and includes a prediction unit that predicts whether or not a fall will occur, and a learning unit that performs supervised learning on the prediction unit. The supervised learning performed by the learning unit includes electronic medical record data of a fall occurrence group, which is a group including hospitalized patients who have been recorded as having fallen or fallen down in the past, and electronic medical record data of a control group, which is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, but are not included in the fall occurrence group. The electronic medical record data of the fall occurrence group includes the data of the fall occurrence group, The fall occurrence prediction device includes attribute data including at least the gender, age, and medical department of the patients included in the fall occurrence group, and the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group, and the prediction unit, in which supervised learning has been performed by the learning unit, predicts whether the prediction target patient will fall or fall within a specified period of hospitalization based on the electronic medical record data of the prediction target patient who is hospitalized, and the electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient, and uses L1 regularized logistic regression as the algorithm of the fall occurrence prediction device.

In one embodiment of the fall prediction device of the present invention, the electronic medical record data of the fall occurrence group includes attribute data including at least the gender, age, and medical department of the patients included in the fall occurrence group, and drug-specific usage data indicating the usage of drugs by the patients included in the fall occurrence group, the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group, and drug-specific usage data indicating the usage of drugs by the patients included in the control group, and the electronic medical record data of the prediction target patient may include attribute data including at least the gender, age, and medical department of the prediction target patient, and drug-specific usage data indicating the usage of drugs by the prediction target patient.

One aspect of the present invention is a method for predicting whether or not a fall will occur, which is when a patient falls or trips during a specified period of hospitalization, comprising a prediction step of predicting whether or not a fall will occur, and a learning step of performing supervised learning before the prediction step is executed, in which the supervised learning used in the learning step includes electronic medical record data of a fall occurrence group, which is a group including hospitalized patients who have been recorded as having fallen or tripped in the past, and electronic medical record data of a control group, which is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, but are not included in the fall occurrence group, and the electronic medical record data of the fall occurrence group includes electronic medical record data of the control group, which is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, but are not included in the fall occurrence group, The electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group, and the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group. In the prediction step, which is executed after the learning step, it is predicted whether or not the prediction target patient will tumble or fall within a specified period of hospitalization based on the electronic medical record data of the prediction target patient who is hospitalized, and the electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient. This is a fall prediction method that uses L1 regularized logistic regression as an algorithm for predicting whether or not the prediction target patient will tumble or fall within the specified period of hospitalization.

One aspect of the present invention is a program for causing a computer to execute a prediction step of predicting whether or not a fall will occur, which is a fall or a fall that occurs during a specified period of time during a patient's hospitalization, and a learning step of performing supervised learning before the prediction step is executed, in which the supervised learning performed in the learning step includes electronic medical record data of a fall occurrence group, which is a group including hospitalized patients who have been recorded as having fallen or fallen in the past, and electronic medical record data of a control group, which is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group and are not included in the fall occurrence group, and the electronic medical record data of the fall occurrence group includes electronic medical record data of the control group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, and The program includes attribute data including at least the gender, age, and medical department of the patients included in the control group, and the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group, and in the prediction step that is executed after the learning step, based on the electronic medical record data of the hospitalized prediction patient, whether or not the prediction patient will tumble or fall within a specified period of hospitalization, the electronic medical record data of the prediction patient includes attribute data including at least the gender, age, and medical department of the prediction patient, and uses L1 regularized logistic regression as an algorithm for predicting whether or not the prediction patient will tumble or fall within the specified period of hospitalization.

One aspect of the present invention is a method for supporting the improvement of medical safety by predicting whether or not a patient in hospital self-removes a line, which is the self-removal of at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body, and the method includes a prediction step for predicting whether or not a patient in hospital self-removes a line, and a learning step for performing supervised learning before the prediction step is performed. The supervised learning used in the learning step includes the electronic medical record data of a self-removal occurrence group, which is a group including hospitalized patients who have been recorded as having self-removed or accidentally removed in the past, and ... including hospitalized patients who match at least the gender, age, and hospital ward of the patients in the self-removal occurrence group, and who are included in the self-removal occurrence group. and electronic medical record data of a control group, which is a group including patients who have not undergone self-removal, the electronic medical record data of the self-removal occurrence group includes attribute data including at least the gender, age, and medical department of the patients included in the self-removal occurrence group, and the electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group, and the prediction step, which is executed after the learning step, predicts whether the prediction target patient will self-remove lines within a predetermined period of hospitalization based on the electronic medical record data of the prediction target patient who is hospitalized, and the electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient.

According to the present invention, it is possible to provide a self-removal prediction device, a self-removal prediction method, and a program that can predict with high accuracy whether or not a hospitalized patient will self-removal of lines, which is the self-removal of at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted inside the body.
In addition, according to the present invention, it is possible to provide a fall/fall occurrence prediction device, a fall/fall occurrence prediction method, and a program that can predict with high accuracy whether or not a fall will occur, which is when a patient falls or trips over a specified period of time during their hospitalization.
In other words, according to the present invention, a medical safety improvement support method that can support the improvement of medical safety can be provided.

This is a diagram to explain the impact of medical accidents such as self-removal of drains or intravenous drips, or falling. FIG. 1 is a diagram showing an example of a flowchart for responding to a fall occurring in a hospital. 1 is a diagram illustrating an example of a self-removal occurrence prediction device 1 according to a first embodiment. FIG. 1 is a diagram showing an example of a specific configuration of a self-removal occurrence prediction device 1 according to a first embodiment. FIG. 4 is a flowchart showing an example of a process executed in the self-removal occurrence prediction device 1 of the first embodiment. FIG. 13 is a diagram illustrating an example of a tipping over/falling occurrence prediction device 2 according to a second embodiment. 13 is a flowchart showing an example of processing executed in a tipping over/falling occurrence prediction device 2 of the second embodiment.

Before describing the embodiments of the self-removal prediction device, self-removal prediction method, program, fall prediction device, fall prediction method, and medical safety improvement support method of the present invention, we will explain the impact of medical accidents such as self-removal of drains or intravenous drips, and falls.

FIG. 1 is a diagram for explaining the impact of medical accidents such as self-removal of drains or intravenous drips, or falls.
As shown in Figure 1, if a medical accident occurs in which a patient removes a drain or an IV drip by themselves, the inpatient who removed the drain or the IV drip by themselves may need to be physically restrained. Also, if a medical accident occurs in which a patient falls, the inpatient who falls may break a bone. As a result, the healthy life expectancy of the inpatient may be shortened.
In addition, when medical accidents occur, such as patients removing drains or IV drips themselves, or falling, medical professionals must respond after the accident occurs. Specifically, medical professionals must evaluate the patient's condition, provide treatment, perform emergency tests, consult with specialists, and write medical safety reports. This can lead to increased burdens on medical professionals and the risk of burnout.

FIG. 2 is a diagram showing an example of a flow chart for responding to a slip and fall that occurs in a hospital.
In the example shown in FIG. 2, when a fall occurs, an initial assessment is then performed, and a request for assistance is made.
After an initial assessment and a request for assistance, basic life support and/or a code blue may be administered.
In addition, after an initial assessment and a request for assistance has been made, an assessment of the patient's condition, including vital signs, injuries, and limitations in range of motion, may be conducted in some cases.
If an abnormality is found as a result of the evaluation of the patient's condition, a request is made for a medical examination to be made by a medical doctor, who then performs an examination.
If the patient condition evaluation reveals no abnormalities, it reports to the medical specialist, who then performs an examination himself or gives instructions to a person who will perform the examination.

The following describes embodiments of the self-removal prediction device, self-removal prediction method, program, fall prediction device, fall prediction method, and medical safety improvement support method of the present invention.

First Embodiment
Fig. 3 is a diagram showing an example of the self-removal occurrence prediction device 1 of the first embodiment. Fig. 4 is a diagram showing an example of a specific configuration of the self-removal occurrence prediction device 1 of the first embodiment.
In the example shown in FIG. 4 , the deep neural network constituting the self-removal occurrence prediction device 1 has an input layer, an output layer, and four hidden layers, but in other examples, the deep neural network constituting the self-removal occurrence prediction device 1 may have an input layer, an output layer, and any number of hidden layers other than four.

In the example shown in FIG. 4, the self-removal prediction device 1 is configured using a deep neural network, but in other examples, the self-removal prediction device 1 may use a machine learning method other than a deep neural network (e.g., L1 regularized logistic regression, random forest, support vector machine, etc.).

3 and 4, self-removal occurrence prediction device 1 predicts whether or not a hospitalized patient will self-removal of at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body. Self-removal occurrence prediction device 1 includes a prediction unit 11 and a learning unit 12.
The prediction unit 11 predicts whether or not self-removal of lines will occur. The learning unit 12 performs supervised learning of the prediction unit 11.
The teacher data used in the supervised learning performed by the learning unit 12 includes electronic medical chart data of the self-removal group and electronic medical chart data of the control group. The self-removal group is a group including hospitalized patients who have been recorded in the past as having self-removed or accidentally removed. The control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the self-removal group, but are not included in the self-removal group.

3 and 4, the electronic medical record data of the voluntary withdrawal group includes attribute data including at least the sex, age, and medical department of patients in the voluntary withdrawal group, and drug-specific usage data showing the usage of drugs by patients in the voluntary withdrawal group. The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of patients in the control group, and drug-specific usage data showing the usage of drugs by patients in the control group.
In another example, drug-specific usage data indicating drug usage by patients included in the voluntary withdrawal group may not be included in the electronic medical record data of the voluntary withdrawal group, and drug-specific usage data indicating drug usage by patients included in the control group may not be included in the electronic medical record data of the control group.

3 and 4, the attribute data of patients in the self-removal group includes sex, age, BMI, height, and medical department, and the attribute data of patients in the control group includes sex, age, BMI, height, and medical department, but in other examples, the attribute data of patients in the self-removal group does not include BMI and/or height, and the attribute data of patients in the control group does not include BMI and/or height. In still other examples, the attribute data of patients in the self-removal group may include items other than sex, age, BMI, height, and medical department (e.g., weight, blood biochemistry data, disease, data on treatment, doctor and nurse charts, vital signs, etc.), and the attribute data of patients in the control group may include items other than sex, age, BMI, height, and medical department (e.g., weight, blood biochemistry data, disease, data on treatment, doctor and nurse charts, vital signs, etc.).

In the examples shown in Figures 3 and 4, the prediction unit 11, which has undergone supervised learning by the learning unit 12, predicts whether the patient to be predicted will remove lines themselves within a specified period of hospitalization based on the electronic medical record data of the patient to be predicted who is currently hospitalized.
In the examples shown in FIG. 3 and FIG. 4, the predetermined period is set to 30 days, but in other examples, the predetermined period may be set to a number of days other than 30 days.

In the examples shown in Figures 3 and 4, the electronic medical record data of the patient to be predicted includes attribute data including at least the gender, age, and medical department of the patient to be predicted, and drug-specific usage data indicating the drug usage by the patient to be predicted.
In other examples (more specifically, examples in which drug-specific usage data indicating drug usage by patients included in the voluntary withdrawal group is not included in the electronic medical record data of the voluntary withdrawal group, and drug-specific usage data indicating drug usage by patients included in the control group is not included in the electronic medical record data of the control group), drug-specific usage data indicating drug usage by the patient to be predicted may not be included in the electronic medical record data of the patient to be predicted.

In the examples shown in Figures 3 and 4, the attribute data of the patient to be predicted includes gender, age, BMI, height, and medical department, but in other examples (more specifically, examples in which the attribute data of patients in the self-removal group do not include BMI and/or height, and the attribute data of patients in the control group do not include BMI and/or height), the attribute data of the patient to be predicted may not include BMI and/or height.
In still other examples (more specifically, examples in which the attribute data of patients in the self-removal group includes data other than gender, age, BMI, height, and medical department, and the attribute data of patients in the control group includes data other than gender, age, BMI, height, and medical department), the attribute data of patients to be predicted may include data other than gender, age, BMI, height, and medical department (for example, weight, blood biochemistry data, data related to diseases and treatments, medical records of doctors and nurses, vital signs, etc.).

In the example shown in FIG. 4, the prediction unit 11, which has undergone supervised learning by the learning unit 12, predicts, based on the electronic medical record data of a hospitalized patient, that there is a 70% probability that the patient will remove lines themselves during the night today, and outputs this as an automatic alert.

In the examples shown in FIGS. 3 and 4, L1 regularized logistic regression is used as the algorithm of the self-removal occurrence prediction device 1.
In another example, an algorithm other than L1 regularized logistic regression (e.g., decision tree, XGBoost, etc.) may be used as the algorithm of the self-removal occurrence prediction device 1.

In the examples shown in Figures 3 and 4, the prediction unit 11 weights the prediction of whether the patient to be predicted will remove lines or other items on his/her own within a specified period differently depending on whether the drug included in the drug-specific usage data indicating the drug usage by the patient to be predicted is a psychotropic drug or not.
Specifically, when the history of psychotropic drug use by the patient to be predicted is included in the drug-specific usage data indicating the drug usage status of the patient to be predicted, the prediction unit 11 predicts a higher likelihood that the patient to be predicted will remove lines on his/her own within a specified period of time than when the history of psychotropic drug use by the patient to be predicted is not included in the drug-specific usage data indicating the drug usage status of the patient to be predicted.

In the examples shown in Figures 3 and 4, the prediction unit 11 predicts that when the patient to be predicted is male, the patient is more likely to remove lines themselves within a specified period of time than when the patient to be predicted is female.
Furthermore, when the age of the patient to be predicted is 65 years or older, the prediction unit 11 predicts that the patient to be predicted is more likely to remove lines by himself within a specified period than when the age of the patient to be predicted is younger than 65 years old.

FIG. 5 is a flowchart showing an example of a process executed in the self-removal occurrence prediction device 1 of the first embodiment.
In the example shown in FIG. 5, in step S11, the learning unit 12 performs supervised learning of the prediction unit 11.
The teacher data used in the supervised learning performed in step S11 includes electronic medical record data of a self-removal group, which is a group including hospitalized patients who have been recorded in the past as having had a self-removal or accidental removal, and electronic medical record data of a control group, which is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients in the self-removal group but are not included in the self-removal group.
The electronic medical record data for the voluntary withdrawal group includes attribute data including at least the gender, age, and medical department of the patients in the voluntary withdrawal group, and drug-specific usage data indicating the usage of drugs by the patients in the voluntary withdrawal group.
The electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group, and drug-specific usage data indicating the usage of drugs by the patients included in the control group.

Next, in step S12, the prediction unit 11 predicts whether or not self-removal of lines will occur.
In other words, in step S12, which is executed after step S11, it is predicted whether the patient to be predicted will remove lines themselves within a specified period of hospitalization based on the electronic medical record data of the hospitalized patient to be predicted.
The electronic medical record data of the patient for whom prediction is to be performed includes attribute data including at least the gender, age, and medical department of the patient for whom prediction is to be performed, and drug-specific usage data indicating the usage of drugs by the patient for whom prediction is to be performed.
That is, in the example shown in FIG. 5, the occurrence of self-removal of lines, i.e., removal of at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted inside the body, by a hospitalized patient is predicted with high accuracy.

Next, the results of research conducted by the inventors regarding the self-removal prediction device 1 of the first embodiment (see Figures 3 and 4) will be described.

[Method]
The present inventors constructed a prediction model for predicting whether or not self-removal of drains and intravenous drips occurred within 30 days of hospitalization in patients aged 65 years or older. A group recorded as self-removal or accidental removal in the hospital incident report was extracted as a self-removal group (delirium occurrence group) as data used for learning the prediction model. A control group was extracted by matching this group with gender, age, and the ward where the drains and intravenous drips occurred. The objective variable of the prediction model was the presence or absence of self-removal, and the feature values were attribute data (gender, age, BMI, height, medical department), and drug usage status at the time of admission. Three types of algorithms were used for the prediction model: L1 regularized logistic regression, decision tree, and XGBoost. The final parameters of the prediction model were determined using Bayesian optimization. The Area Under Receiver Operating Curve (AUROC) was used as the evaluation index of the prediction model. An 8-fold cross-validation was used for evaluation. Python 3.8.1 was used for all analyses.

[Application of the results]
A prediction model was constructed to predict whether or not self-removal of drains and drips occurred within 30 days of hospitalization for a total of approximately 3,000 cases, with 783 cases extracted from the group with self-removal of drains and drips and 2,570 cases extracted from the control group. When only attribute data was used, the AUROC was 0.616 for L1 regularized logistic regression, 0.882 for decision tree, and 0.988 for XGBoost. When the drug usage status at the time of hospitalization was added, the AUROC was 0.723 for L1 regularized logistic regression, 0.707 for decision tree, and 0.986 for XGBoost. Of these, the AUROC values for decision tree and XGBoost were higher than expected, and there is a possibility of overlearning, so L1 regularized logistic regression will be used for future studies.
The predictive model developed in this study makes it possible to predict the risk of hospitalized patients removing drains or intravenous drips themselves.

In other words, the self-removal prediction device 1 of the first embodiment can help improve medical safety by predicting whether or not a hospitalized patient will remove at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body by himself/herself.

The self-removal occurrence prediction device 1 of the first embodiment can predict the occurrence of self-removal in a hospital with higher accuracy than conventional technology by using the patient's electronic medical record data and artificial intelligence technology. As a result, it can contribute to reducing medical expenses and extending the healthy life expectancy of the elderly.

Second Embodiment
FIG. 6 is a diagram showing an example of a tipping over and falling occurrence predicting device 2 according to the second embodiment.
The fall prediction device 2 of the second embodiment can be configured by a deep neural network, similar to the self-removal prediction device 1 shown in Fig. 4. Alternatively, the fall prediction device 2 may use a machine learning method other than the deep neural network (e.g., L1 regularized logistic regression, random forest, support vector machine, etc.).

6, the fall occurrence prediction device 2 predicts whether or not a fall will occur, that is, a patient will fall or slip down a certain distance during a predetermined period of hospitalization. The fall occurrence prediction device 2 includes a prediction unit 21 and a learning unit 22.
The prediction unit 21 predicts whether or not a fall will occur. The learning unit 22 performs supervised learning of the prediction unit 21.
The teacher data used in the supervised learning performed by the learning unit 22 includes electronic medical record data of the fall occurrence group and electronic medical record data of the control group. The fall occurrence group is a group including hospitalized patients who have been recorded as having fallen or fallen in the past. The control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, but are not included in the fall occurrence group.

6, the electronic medical record data of the fall occurrence group includes attribute data including at least the sex, age, and medical department of the patients in the fall occurrence group, and drug-specific use data showing the use of drugs by the patients in the fall occurrence group. The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of the patients in the control group, and drug-specific use data showing the use of drugs by the patients in the control group.
In another example, drug-specific usage data indicating the usage of drugs by patients included in the fall occurrence group may not be included in the electronic medical record data of the fall occurrence group, and drug-specific usage data indicating the usage of drugs by patients included in the control group may not be included in the electronic medical record data of the control group.

In the example shown in FIG. 6, the attribute data of patients in the fall occurrence group includes sex, age, BMI, height, and medical department, and the attribute data of patients in the control group includes sex, age, BMI, height, and medical department, but in other examples, the attribute data of patients in the fall occurrence group does not include BMI and/or height, and the attribute data of patients in the control group does not include BMI and/or height. In still other examples, the attribute data of patients in the fall occurrence group may include things other than sex, age, BMI, height, and medical department (e.g., weight, blood biochemistry data, disease, data on treatment, doctor and nurse chart records, vital signs, etc.), and the attribute data of patients in the control group may include things other than sex, age, BMI, height, and medical department (e.g., weight, blood biochemistry data, disease, data on treatment, doctor and nurse chart records, vital signs, etc.).

In the example shown in Figure 6, the prediction unit 21, which has undergone supervised learning by the learning unit 22, predicts whether the patient to be predicted will tumble or fall within a specified period of hospitalization based on the electronic medical record data of the patient to be predicted who is currently hospitalized.
In the example shown in FIG. 6, the predetermined period is set to 30 days, but in other examples, the predetermined period may be set to a number of days other than 30 days.

In the example shown in Figure 6, the electronic medical record data of the patient to be predicted includes attribute data including at least the gender, age, and medical department of the patient to be predicted, and drug-specific usage data indicating the drug usage by the patient to be predicted.
In other examples (more specifically, examples in which drug-specific usage data indicating drug usage by patients included in the fall occurrence group is not included in the electronic medical record data of the fall occurrence group, and drug-specific usage data indicating drug usage by patients included in the control group is not included in the electronic medical record data of the control group), drug-specific usage data indicating drug usage by the patient to be predicted may not be included in the electronic medical record data of the patient to be predicted.

In the example shown in FIG. 6 , the attribute data of the patient to be predicted includes gender, age, BMI, height, and medical department, however, in other examples (more specifically, examples in which the attribute data of patients in the fall occurrence group does not include BMI and/or height, and the attribute data of patients in the control group does not include BMI and/or height), the attribute data of the patient to be predicted may not include BMI and/or height.
In yet another example (more specifically, an example in which the attribute data of patients in the fall occurrence group includes data other than gender, age, BMI, height, and medical specialty, and the attribute data of patients in the control group includes data other than gender, age, BMI, height, and medical specialty), the attribute data of the patient to be predicted may include data other than gender, age, BMI, height, and medical specialty (e.g., weight, blood biochemistry data, data regarding diseases and treatments, medical records of doctors and nurses, vital signs, etc.).

In the example shown in FIG. 6, L1 regularized logistic regression is used as the algorithm of the fall occurrence prediction device 2.

In the example shown in Figure 6, the prediction unit 21 weights the prediction of whether the patient to be predicted will collapse or fall within a specified period differently depending on whether the drug included in the drug-specific usage data indicating the drug usage by the patient to be predicted is a psychotropic drug.
Specifically, when the history of psychotropic drug use by the patient to be predicted is included in the drug-specific usage data indicating the drug usage status of the patient to be predicted, the prediction unit 21 predicts a higher likelihood that the patient to be predicted will have a fall or a drop within a specified period of time than when the history of psychotropic drug use by the patient to be predicted is not included in the drug-specific usage data indicating the drug usage status of the patient to be predicted.

In the example shown in Figure 6, the prediction unit 21 predicts that the patient to be predicted will be more likely to fall or slip within a specified period of time when the patient to be predicted is male than when the patient to be predicted is female.
Furthermore, when the age of the patient to be predicted is 65 years or older, the prediction unit 21 predicts that the patient to be predicted is more likely to fall or slip within a specified period of time than when the age of the patient to be predicted is younger than 65 years old.

FIG. 7 is a flowchart showing an example of a process executed in the tipping over/falling occurrence predicting device 2 of the second embodiment.
In the example shown in FIG. 7, in step S21, the learning unit 22 performs supervised learning of the prediction unit 21.
The teacher data used in the supervised learning performed in step S21 includes electronic medical record data of a fall occurrence group, which is a group including hospitalized patients who have been recorded as having fallen or fallen in the past, and electronic medical record data of a control group, which is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, but are not included in the fall occurrence group.
The electronic medical record data of the fall occurrence group includes attribute data including at least the gender, age, and medical specialty of the patients included in the fall occurrence group, and drug-specific usage data indicating the usage of drugs by the patients included in the fall occurrence group.
The electronic medical record data of the control group includes attribute data including at least the gender, age, and medical department of the patients included in the control group, and drug-specific usage data indicating the usage of drugs by the patients included in the control group.

Next, in step S22, the prediction unit 21 predicts whether or not a fall will occur.
In other words, in step S22, which is executed after step S21, it is predicted whether or not the patient to be predicted will fall or slip within a specified period of hospitalization based on the electronic medical record data of the hospitalized patient to be predicted.
The electronic medical record data of the patient for whom prediction is to be performed includes attribute data including at least the gender, age, and medical department of the patient for whom prediction is to be performed, and drug-specific usage data indicating the usage of drugs by the patient for whom prediction is to be performed.
7, an L1 regularized logistic regression is used as an algorithm for predicting whether a patient to be predicted will fall or drop within a predetermined period of hospitalization. Therefore, overlearning of the prediction unit 21 can be suppressed.
That is, in the example shown in FIG. 7, the occurrence or nonoccurrence of a fall, that is, a fall or tripping over a floor, of a patient during a specified period of hospitalization is predicted with high accuracy.

Next, we will explain the results of research conducted by the inventors regarding the second embodiment of the fall prediction device 2 (see Figure 6).

[Method]
The present inventors constructed a prediction model for predicting whether or not a fall occurred within 30 days of hospitalization for patients aged 65 years or older. A group recorded as a fall or fall in the hospital incident report was extracted as a fall occurrence group as data used to train the prediction model. A control group was extracted by matching this group with gender, age, and the ward where the fall occurred. The objective variable of the prediction model was the presence or absence of a fall, and the attribute data (gender, age, BMI, height, medical department) and the drug usage status at the time of admission were set as features. Three types of algorithms were used for the prediction model: L1 regularized logistic regression, decision tree, and XGBoost. The final parameters of the prediction model were determined using Bayesian optimization. AUROC was used as the evaluation index of the prediction model. Eight-fold cross-validation was used for evaluation. Python 3.8.1 was used for all analyses.

[Application of the results]
A prediction model was constructed to predict whether or not falls occurred within 30 days of hospitalization, with 834 cases in the fall occurrence group and 3,379 cases in the control group, for a total of approximately 4,000 cases. When only attribute data was used, the AUROC was 0.607 for L1 regularized logistic regression, 0.861 for decision tree, and 0.986 for XGBoost. When the drug usage status at the time of hospitalization was added, the AUROC was 0.649 for L1 regularized logistic regression, 0.788 for decision tree, and 0.986 for XGBoost. Of these, the AUROC values for decision tree and XGBoost were higher than expected, and there is a possibility of overlearning, so L1 regularized logistic regression will be used for future studies.
The predictive model developed in this study makes it possible to predict the risk of falls among hospitalized patients.

In other words, the fall prediction device 2 of the second embodiment can help improve medical safety by predicting whether or not a fall will occur, which is when a patient falls or trips down a certain period of time during their hospitalization.

The second embodiment of the fall prediction device 2 uses electronic medical record data of patients and artificial intelligence technology to predict whether or not a fall will occur within a hospital with greater accuracy than conventional technology. As a result, it can contribute to reducing medical costs and extending the healthy lifespan of the elderly.

Although the above describes the form for carrying out the present invention using the embodiments, the present invention is not limited to these embodiments in any way, and various modifications and substitutions can be made without departing from the scope of the present invention. The configurations described in the above-mentioned embodiments and examples may be combined.

In addition, all or part of the functions of each unit of the self-removal occurrence prediction device 1 or the fall occurrence prediction device 2 in the above-mentioned embodiment may be realized by recording a program for realizing these functions in a computer-readable recording medium, reading the program recorded in the recording medium into a computer system, and executing it. Note that the "computer system" referred to here includes hardware such as the OS and peripheral devices.
Furthermore, "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, and storage units such as hard disks built into computer systems. Furthermore, "computer-readable recording medium" may also include those that dynamically hold a program for a short period of time, such as a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line, and those that hold a program for a certain period of time, such as a volatile memory inside a computer system that serves as a server or client in such cases. Furthermore, the above program may be one that realizes part of the above-mentioned functions, or may be one that can realize the above-mentioned functions in combination with a program already recorded in the computer system.

1... Self-removal occurrence prediction device, 11... Prediction unit, 12... Learning unit, 2... Fall occurrence prediction device, 21... Prediction unit, 22... Learning unit

Claims (11)

A self-removal occurrence prediction device for predicting whether or not a hospitalized patient will self-remove at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body, the device comprising:
A prediction unit for predicting whether or not self-removal of lines will occur;
A learning unit that performs supervised learning of the prediction unit,
The teacher data used in the supervised learning performed by the learning unit includes:
Electronic medical record data from the self-removal group, which included hospitalized patients who had previously been recorded as having self-removal or accidental removal, and
The electronic medical record data of a control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the self-removal occurrence group, but are not included in the self-removal occurrence group,
The electronic medical record data of the self-removal occurrence group includes attribute data including at least the sex, age, and medical department of patients included in the self-removal occurrence group,
The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of the patients included in the control group;
The prediction unit, in which supervised learning has been performed by the learning unit, predicts whether the prediction target patient will remove lines or the like within a predetermined period of hospitalization based on electronic medical record data of the prediction target patient who is hospitalized,
The electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient.
Self-removal prediction device. The electronic medical record data for the self-removal group includes:
Attribute data including at least the gender, age, and medical department of patients included in the self-removal occurrence group;
The self-removal occurrence group includes drug-specific usage status data showing the usage status of drugs by patients included in the self-removal occurrence group,
The electronic medical record data of the control group includes:
Attribute data including at least the gender, age, and medical department of patients included in the control group;
and drug-specific usage data showing drug usage by patients included in the control group;
The electronic medical record data of the prediction target patient includes:
Attribute data including at least the gender, age, and medical department of the patient to be predicted;
The prediction target patient's drug usage status data is included.
The self-removal occurrence prediction device according to claim 1. The self-removal occurrence prediction device uses L1 regularized logistic regression as an algorithm.
The self-removal occurrence prediction device according to claim 2. The prediction unit is
weighting the prediction as to whether the patient to be predicted will remove lines or the like within the predetermined period of time differently depending on whether a drug included in drug-specific use status data indicating the use status of drugs by the patient to be predicted is a psychotropic drug or not;
The self-removal occurrence prediction device according to claim 2. A method for predicting whether or not a patient in a hospital self-removal of a line, which is at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body, will occur, the method comprising:
A prediction step of predicting whether or not self-removal of lines will occur;
A learning step of performing supervised learning before the prediction step is performed,
The training data used in the supervised learning performed in the learning step includes:
Electronic medical record data from the self-removal group, which included hospitalized patients who had previously been recorded as having self-removal or accidental removal, and
The electronic medical record data of a control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the self-removal occurrence group, but are not included in the self-removal occurrence group,
The electronic medical record data of the self-removal occurrence group includes attribute data including at least the sex, age, and medical department of patients included in the self-removal occurrence group,
The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of the patients included in the control group;
In the prediction step, which is executed after the learning step is executed, it is predicted whether or not the prediction target patient will remove lines or the like by himself/herself within a predetermined period of hospitalization based on electronic medical record data of the prediction target patient who is hospitalized,
The electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient.
A method for predicting the occurrence of self-removal. On the computer,
A prediction step of predicting whether or not a patient in a hospital self-removal of a line, which is the self-removal of at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body, will occur;
A learning step of performing supervised learning before the prediction step is executed,
The training data used in the supervised learning performed in the learning step includes:
Electronic medical record data from the self-removal group, which included hospitalized patients who had previously been recorded as having self-removal or accidental removal, and
The electronic medical record data of a control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the self-removal occurrence group, but are not included in the self-removal occurrence group,
The electronic medical record data of the self-removal occurrence group includes attribute data including at least the sex, age, and medical department of patients included in the self-removal occurrence group,
The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of the patients included in the control group;
In the prediction step, which is executed after the learning step is executed, it is predicted whether or not the prediction target patient will remove lines or the like by himself/herself within a predetermined period of hospitalization based on electronic medical record data of the prediction target patient who is hospitalized,
The electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient.
program. A fall occurrence prediction device for predicting whether or not a fall will occur, which is a fall or a fall of a patient during a predetermined period of hospitalization, comprising:
A prediction unit that predicts whether or not a fall will occur;
A learning unit that performs supervised learning of the prediction unit,
The teacher data used in the supervised learning performed by the learning unit includes:
Electronic medical record data from the fall occurrence group, which is a group that includes hospitalized patients who had a history of falls.
The electronic medical record data of a control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, but are not included in the fall occurrence group,
The electronic medical record data of the fall occurrence group includes attribute data including at least the gender, age, and medical department of the patients included in the fall occurrence group,
The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of the patients included in the control group;
The prediction unit, after the supervised learning by the learning unit, predicts whether or not the prediction target patient will tumble or fall within a predetermined period of hospitalization based on electronic medical record data of the prediction target patient who is hospitalized,
The electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient,
The fall prediction device uses L1 regularized logistic regression as an algorithm.
A device for predicting the occurrence of falls. The electronic medical record data of the fall occurrence group includes:
Attribute data including at least the gender, age, and medical department of patients included in the fall occurrence group;
and drug-specific use status data showing the use status of drugs by patients included in the fall occurrence group,
The electronic medical record data of the control group includes:
Attribute data including at least the gender, age, and medical department of patients included in the control group;
and drug-specific usage data showing drug usage by patients included in the control group;
The electronic medical record data of the prediction target patient includes:
Attribute data including at least the gender, age, and medical department of the patient to be predicted;
The prediction target patient's drug usage status data is included.
The device for predicting the occurrence of a fall according to claim 7. A method for predicting whether or not a patient will fall during a predetermined period of hospitalization, comprising:
A prediction step of predicting whether or not a fall will occur;
A learning step of performing supervised learning before the prediction step is performed,
The training data used in the supervised learning performed in the learning step includes:
Electronic medical record data from the fall occurrence group, which is a group that includes hospitalized patients who had a history of falls.
The electronic medical record data of a control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, but are not included in the fall occurrence group,
The electronic medical record data of the fall occurrence group includes attribute data including at least the gender, age, and medical department of the patients included in the fall occurrence group,
The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of the patients included in the control group;
In the prediction step, which is executed after the learning step is executed, it is predicted whether or not the prediction target patient will tumble or fall within a predetermined period of hospitalization based on electronic medical record data of the prediction target patient who is hospitalized,
The electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient,
An L1 regularized logistic regression is used as an algorithm for predicting whether the prediction target patient will fall or fall within the predetermined period of hospitalization;
A method for predicting the occurrence of falls. On the computer,
A prediction step of predicting whether or not a fall will occur, that is, a fall or a drop in a predetermined period during a patient's hospitalization;
A learning step of performing supervised learning before the prediction step is executed,
The training data used in the supervised learning performed in the learning step includes:
Electronic medical record data from the fall occurrence group, which is a group that includes hospitalized patients who had a history of falls.
The electronic medical record data of a control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the fall occurrence group, but are not included in the fall occurrence group,
The electronic medical record data of the fall occurrence group includes attribute data including at least the gender, age, and medical department of the patients included in the fall occurrence group,
The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of the patients included in the control group;
In the prediction step, which is executed after the learning step is executed, it is predicted whether or not the prediction target patient will tumble or fall within a predetermined period of hospitalization based on electronic medical record data of the prediction target patient who is hospitalized,
The electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient,
An L1 regularized logistic regression is used as an algorithm for predicting whether the prediction target patient will fall or fall within the predetermined period of time during hospitalization.
program. A medical safety improvement support method for supporting improvement of medical safety by predicting whether or not a hospitalized patient will self-remove at least one of a drain, an IV needle, a blood transfusion needle, a catheter, and a tube inserted into the body, the method comprising:
A prediction step of predicting whether or not self-removal of lines will occur;
A learning step of performing supervised learning before the prediction step is performed,
The training data used in the supervised learning performed in the learning step includes:
Electronic medical record data from the self-removal group, which included hospitalized patients who had previously been recorded as having self-removal or accidental removal, and
The electronic medical record data of a control group is a group including hospitalized patients who match at least the gender, age, and hospital ward of the patients included in the self-removal occurrence group, but are not included in the self-removal occurrence group,
The electronic medical record data of the self-removal occurrence group includes attribute data including at least the sex, age, and medical department of patients included in the self-removal occurrence group,
The electronic medical record data of the control group includes attribute data including at least the sex, age, and medical department of the patients included in the control group;
In the prediction step, which is executed after the learning step is executed, it is predicted whether or not the prediction target patient will remove lines or the like by himself/herself within a predetermined period of hospitalization based on electronic medical record data of the prediction target patient who is hospitalized,
The electronic medical record data of the prediction target patient includes attribute data including at least the gender, age, and medical department of the prediction target patient.
Methods to help improve medical safety.

Images