Method and apparatus for predicting bed occupancy rate

Abstract

Disclosed are an apparatus and a method for predicting a bed occupancy rate in a hospital. The method for predicting a bed occupancy rate, performed by the apparatus for predicting a bed occupancy rate, according to an embodiment, may comprise the steps of: periodically collecting bed information data including information indicating whether a patient occupies each bed and hospital room identification information indicating a corresponding bed; generating, from the collected bed information data, ward data based on patient occupancy information dependent on the time in a ward including each bed; generating, from the collected bed information data, hospital room data based on at least one of patient occupancy information dependent on the time in a hospital room including each bed or time-independent hospital room identification information; training a machine learning-based occupancy rate prediction model by using training data generated on the basis of dividing at least one of the ward data and time-dependent information among the hospital room data into units of predetermined time periods; and applying the trained machine learning-based occupancy rate prediction model to input bed information data, so as to output occupancy rate information for at least one of a ward or a hospital room.

Description

Method and device for predicting hospital bed occupancy

[0001] Below, we describe a technique for predicting patient occupancy rates for hospital beds.

[0002] In today's healthcare system, there is growing interest in efficiently utilizing limited resources. Healthcare resources can be broadly divided into three categories: human resources, physical capital, and consumables. The appropriate and optimal use of these resources is crucial for improving the quality of care and managing more patients. Hospital beds, a physical resource within the healthcare system, are used for rest, hospitalization, and post-operative recovery, and are a key factor directly impacting patient satisfaction within the hospital. Due to limited space, hospitals have a limited number of beds. Because the number and function of beds are difficult to change, they are often fixed for budgetary or environmental reasons. Monitoring and predicting bed occupancy rates to monitor hospital capacity can impact hospital operational planning. Simply high bed occupancy rates can negatively impact staff health and increase the risk of infection. Therefore, simply focusing on maintaining high bed occupancy rates may not necessarily be beneficial to hospitals. Therefore, understanding planned patient capacity is crucial. However, assessing and predicting bed occupancy through human resources has limitations. Therefore, a technology is required to secure data on bed occupancy, appropriately preprocess the acquired data to create or train a model to predict bed occupancy, and predict bed occupancy using the model.

[0003] A method for predicting a hospital bed occupancy rate performed by a processor according to one embodiment may include: periodically collecting hospital bed information data including whether a patient occupies each bed and hospital room identification information indicating the corresponding bed; generating ward data based on time-dependent patient occupancy information in a ward including each bed from the collected hospital bed information data; generating hospital room data based on at least one of time-dependent patient occupancy information or time-independent hospital room identification information in a ward including each bed from the collected hospital bed information data; training a machine learning-based occupancy rate prediction model using training data generated based on dividing at least one of time-dependent information and the ward data in the ward data into predetermined time periods; and outputting occupancy information of at least one of a ward or a hospital room by applying the trained machine learning-based occupancy rate prediction model to input hospital bed information data.

[0004] The step of generating the above ward data may include a step of determining ward resource information and ward occupancy rate for each ward including each bed from a result of combining the collected bed information data and the point-in-time data that distinguishes the point-in-time of collection of the bed information data, and the point-in-time data may include information indicating whether the collection point-in-time falls on at least one of a weekend or a public holiday.

[0005] The step of generating the above ward data may include a step of extracting data indicating whether a patient can be admitted to a ward including each bed from the collected bed information data; and a step of generating a bed occupancy value based on ward resource information and ward capacity information for each ward.

[0006] The step of generating the above-mentioned hospital room data includes a step of determining hospital room resource information and hospital room occupancy rate for each hospital room including each hospital bed from a result of combining the collected hospital bed information data and the time point data that distinguishes the time point of collection of the hospital bed information data, and the time point data may include information indicating whether the collection time point falls on at least one of a weekend or a public holiday.

[0007] The step of generating the above-mentioned hospital room data may include a step of extracting data indicating whether a patient can be admitted to a hospital room including each hospital bed from the collected hospital bed information data; and a step of generating a hospital room occupancy rate value based on the hospital room occupancy information and the hospital room occupancy information for each hospital room.

[0008] The step of generating the above-mentioned hospital room data may include the step of dividing the time-independent hospital room identification information into first identification information indicating the corresponding hospital room and second identification information other than the first identification information; the step of transforming the separated first identification information and the second identification information into predetermined numerical values; and the step of combining the predetermined numerical value with data indicating whether the hospital room is available for patient admission, the hospital room capacity information, the hospital room resource information, and the hospital room occupancy value, and the time-independent hospital room identification information may include information indicating a non-emergency of the patient during a period in which the hospital bed information data is periodically collected.

[0009] The step of training the machine learning-based occupancy prediction model may include a step of generating training data based on dividing the ward data into at least one period unit of 7 days or 30 days.

[0010] The step of training the machine learning-based occupancy prediction model may include a step of generating training data based on dividing time-dependent information in the hospital room data into at least one period unit of 3 days or 7 days.

[0011] The step of outputting occupancy information of at least one of the ward or the ward room may include: extracting a first output including time-series information of the dynamic data by applying a first model including an LSTM layer (long short term memory layer) to dynamic data generated based on time-dependent information in the ward data and dividing at least one of the ward data into predetermined time periods; extracting a second output including features of the static data by applying a second model including a dense layer to static data generated based on time-independent ward identification information in the ward data; and outputting occupancy information of at least one of the ward or the ward for each collection time point of the ward information data based on a concatenation of the first output and the second output.

[0012] According to one embodiment, a bed occupancy prediction device may include a processor that periodically collects bed information data including whether a bed is occupied by a patient and bed identification information indicating the corresponding bed, generates ward data based on time-dependent patient occupancy information in a ward including each bed from the collected bed information data, generates hospital room data based on at least one of time-dependent patient occupancy information or time-independent hospital room identification information in a ward including each bed from the collected bed information data, trains a machine learning-based occupancy prediction model using training data generated based on dividing at least one of the time-dependent information and the ward data in the ward data into predetermined time periods, and applies the trained machine learning-based occupancy prediction model to input bed information data, thereby outputting occupancy information of at least one of a ward or a hospital room.

[0013] The processor can generate the ward data by determining ward resource information and ward occupancy rate for each ward including each bed from a result of combining the point-in-time data that distinguishes the point-in-time of collection of the ward information data and the collected ward information data, and the point-in-time data can include information indicating whether the point-in-time of collection falls on at least one of a weekend or a public holiday.

[0014] The processor can generate the ward data by extracting data indicating whether a patient can be admitted to a ward including each bed from the collected bed information data, and generating a bed occupancy value based on ward resource information and ward capacity information for each ward.

[0015] The processor can generate the hospital room data by determining hospital room resource information and hospital room occupancy rate for each hospital room including each hospital bed from a result of combining the collected hospital bed information data and the time point data that distinguishes the time point of collection of the hospital bed information data, and the time point data can include information indicating whether the collection time point falls on at least one of a weekend or a public holiday.

[0016] The processor can generate the hospital room data by extracting data indicating whether a patient can be admitted to a hospital room including each hospital bed from the collected hospital bed information data, and generating a hospital room occupancy rate value based on the hospital room resource information and the hospital room resource information for each hospital room.

[0017] The processor may generate the hospital room data by separating the time-independent hospital room identification information into first identification information indicating the corresponding hospital room and second identification information other than the first identification information, transforming the separated first identification information and the second identification information into predetermined numerical values, and combining the predetermined numerical values ​​with data indicating whether the hospital room can accommodate a patient, the hospital room capacity information, the hospital room resource information, and the hospital room occupancy rate value, and the time-independent hospital room identification information may include information indicating a non-emergency of the patient during a period in which the hospital bed information data is periodically collected.

[0018] The processor can train the machine learning-based occupancy prediction model by generating training data based on dividing the ward data into at least one period unit of 7 days or 30 days.

[0019] The processor can train the machine learning-based occupancy prediction model by generating training data based on dividing time-dependent information from the ward data into at least one period unit of 3 days or 7 days.

[0020] The processor extracts a first output including time-series information of the dynamic data by applying a first model including an LSTM layer (long short term memory layer) to dynamic data generated based on dividing at least one of time-dependent information and the ward data from the ward data into predetermined time periods, and extracts a second output including features of the static data by applying a second model including a dense layer to static data generated based on time-independent ward identification information from the ward data, and outputs occupancy information of at least one of the ward or ward based on a concatenation of the first output and the second output, by outputting occupancy information of at least one of the ward or ward for each collection time point of the ward information data.

[0021] Figure 1 illustrates a bed occupancy prediction device according to one embodiment.

[0022] FIG. 2 illustrates a flowchart of a method for predicting bed occupancy in a hospital by a bed occupancy prediction device according to one embodiment.

[0023] FIG. 3 is a block diagram of a bed occupancy prediction operation of a bed occupancy prediction device according to one embodiment.

[0024] FIG. 4 illustrates a bed occupancy prediction device according to one embodiment training a machine learning-based occupancy prediction model based on different training data.

[0025] FIG. 5 illustrates a machine learning-based occupancy prediction model architecture according to one embodiment.

[0026] FIGS. 6A and 6B illustrate user interface displays for hospital bed occupancy prediction results predicted by a bed occupancy prediction device according to one embodiment.

[0027] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Therefore, the actual implementation is not limited to the specific embodiments disclosed, and the scope of this specification includes modifications, equivalents, or alternatives within the technical concepts described in the embodiments.

[0028] Although terms such as "first" or "second" may be used to describe various components, these terms should be interpreted solely to distinguish one component from another. For example, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component.

[0029] When it is said that a component is "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but there may also be other components in between.

[0030] Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "has" should be understood to indicate the presence of a described feature, number, step, operation, component, part, or combination thereof, but not to exclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0031] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art. Terms defined in commonly used dictionaries should be interpreted to have a meaning consistent with their meaning in the context of the relevant technology, and will not be interpreted in an idealized or overly formal sense unless explicitly defined herein.

[0032] Hereinafter, embodiments will be described in detail with reference to the attached drawings. In the description with reference to the attached drawings, identical components are assigned the same reference numerals regardless of the drawing numbers, and redundant descriptions thereof will be omitted.

[0033] Fig. 1 illustrates a bed occupancy prediction device (100) according to one embodiment. The bed occupancy prediction device (100) can distinguish between beds occupied by patients and unoccupied beds in a hospital and predict the patient occupancy rate. For example, the bed occupancy prediction device (100) periodically collects information indicating whether there is a patient occupying a bed, and based on the collected information, can predict the patient occupancy rate of a bed, and further, a ward or ward including the bed. For example, the bed occupancy prediction device (100) can collect bed information data collected every hour and predict the patient occupancy rate of a bed based on the data.

[0034] For reference, bed information data may include, but is not limited to, the date the data was collected, information on the ward and room containing the bed, whether the patient is in the bed, the bed cleaning status, and detailed information for each room.

[0035] A bed occupancy prediction device (100) according to one embodiment may include a memory (110) and a processor (120).

[0036] The memory (110) can store a machine learning-based occupancy prediction model for predicting bed occupancy, computer-executable instructions for training the occupancy prediction model, parameters of the occupancy prediction model (e.g., weights), and bed information data.

[0037] The processor (120) may periodically collect bed information data, including whether a bed is occupied by a patient and room identification information indicating the corresponding bed. For reference, the room identification information may be information about a room containing a bed determined to be occupied by a patient. For example, the room identification information may be time-independent information, such as the room class code of the room determined to be occupied by a patient, whether it is a nuclear medicine ward, whether it is an aseptic room, or whether it is a special room. The room identification information is described in detail in FIG. 2 below.

[0038] The processor (120) may collect bed information data at predetermined time intervals. For example, the processor (120) may collect bed information data including whether each bed in the hospital is occupied by a patient and identification information for the bed containing the bed determined to be occupied by a patient at one-hour intervals. The bed information data collected by the processor (120) may be stored in the memory (110) with each category as a column. For example, each category of the bed information data collected by the processor (120) may include the date the data was collected, whether the patient is in the bed, the bed cleaning status, information about the ward to which the bed belongs, and information about the ward to which the bed belongs. The bed information data may be data collected at hourly intervals by arranging the categories of information included in the bed information data into a single column. The processor (120) may distinguish the date of the bed information data by adding data indicating weekends or holidays to the bed information data. Data indicating a weekend or public holiday by the processor (120) is specifically described in FIG. 3 below.

[0039] The processor (120) may generate ward data based on patient occupancy information that depends on time in the ward where each bed is included from the collected bed information data. For example, the processor (120) may generate ward data based on bed information data included in a specific ward among the collected bed information data. For example, the bed information data collected by the processor (120) may include data on beds belonging to an 'internal medicine ward', data on beds belonging to a 'surgical ward', or data on beds belonging to a 'pediatric ward'. The processor (120) may generate internal medicine ward data based on data on beds belonging to an 'internal medicine ward' among the collected bed information data. The processor (120) may generate surgical ward data based on data on beds belonging to a 'surgical ward'. While the above example illustrates the generation of ward data for each ward by the processor (120), the processor (120) may also simultaneously generate multiple ward data corresponding to each of multiple wards. The ward data generated by the processor (120) may include information regarding the ward abbreviation, ward capacity, ward resources, ward occupancy rate, and the date the ward data was collected. The ward data is described in detail below in FIG. 2.

[0040] The processor (120) can generate hospital room data based on at least one of time-dependent patient occupancy information or time-independent hospital room identification information in a hospital room containing each bed from the collected hospital bed information data. The method by which the processor (120) generates hospital room data based on time-dependent patient occupancy information in a hospital room containing each bed from the collected hospital bed information data is the same as the method for generating hospital bed data above, and thus will not be repeated. The processor (120) can generate hospital room data based on time-independent hospital room identification information. For example, the time-independent hospital room identification information may include information for identifying a hospital room containing a corresponding bed, regardless of the period in which the hospital bed information data was collected. For example, the time-independent hospital room identification information may include the floor of the hospital room to which the bed determined to be occupied by a patient by the processor (120) belongs, the room number of the corresponding hospital room, the grade code of the corresponding hospital room, etc. The hospital room data generated by the processor (120) will be described in detail with reference to FIG. 2.

[0041] The processor (120) may train a machine learning-based occupancy prediction model using training data generated based on dividing at least one of time-dependent information and ward data from the ward data into predetermined time-period units. The processor (120) may divide at least one of the time-dependent information and ward data from the ward data into data bundles of predetermined time-period units through windows. For example, the processor (120) may generate training data based on dividing the ward data into 7-day units through windows that divide the data into 7-day units. Each 7-day unit of ward data may include information indicating whether a bed in the ward is occupied by a patient for a week. As another example, the processor (120) may generate training data based on dividing the ward data into 3-day units through windows that divide the data into 3-day units. Each 3-day unit of ward data may include information on changes in patient occupancy for a bed for a shorter period of time compared to each 7-day unit of ward data.

[0042] The processor (120) can train a machine learning-based occupancy prediction model using the generated training data. For reference, the machine learning-based occupancy prediction model can include a neural network including multiple layers, etc. The occupancy prediction model can map input data and output data to each other through supervised learning using a machine learning technique. In the case of supervised learning, the machine learning-based occupancy prediction model described above can be trained based on a training data set including pairs of training inputs (e.g., ward data or ward room data for training) and training outputs mapped to the training inputs (e.g., ground truth values ​​classified by an expert as values ​​representing the occupancy of hospital beds for the ward data or ward room data for training). For example, the machine learning-based occupancy prediction model can be trained to output training outputs from training inputs. The model during training can generate temporary outputs in response to the training inputs, and can be trained such that loss between the temporary outputs and the training outputs (e.g., ground truth values) is minimized. During the training process, parameters of a machine learning-based occupancy prediction model (e.g., connection weights between multiple layers) may be updated based on the loss. However, this training method is merely an example, and the methods for training the occupancy prediction model are not limited to this.

[0043] For reference, the proposed machine learning-based market share prediction model may include models that can simultaneously input time-dependent information (e.g., dynamic data) as well as time-independent information (e.g., static data). For example, the machine learning-based market share prediction model may include a model that combines a first model that inputs dynamic data and a second model that inputs static data. A specific machine learning-based market share prediction model is described in detail in FIG. 5 below.

[0044] The processor (120) can output occupancy information for at least one ward or room by applying a trained machine learning-based occupancy prediction model to input bed information data. For example, the processor (120) can output occupancy prediction information for at least one ward or room at a 'desired point in time' by applying a trained machine learning-based occupancy prediction model to 'current' bed information data.

[0045]

[0046] FIG. 2 illustrates a flowchart of a method for predicting bed occupancy in a hospital by a bed occupancy prediction device according to one embodiment.

[0047] In step (210), a bed occupancy prediction device according to one embodiment may periodically collect bed information data including whether a bed is occupied by a patient and bed identification information indicating the corresponding bed. For example, the bed occupancy prediction device may collect bed occupancy by a patient, bed cleaning status, and data collection time information (e.g., year, month, day, hour, day of the week, etc.) at one-hour intervals. At the same time, the bed occupancy prediction device may collect bed identification information for a bed that includes the corresponding bed. For example, the bed occupancy prediction device may collect data substituted with 1 if the bed is occupied by a patient, and data substituted with 0 if the bed is not occupied by the patient. For example, assuming that there are a total of 6 beds in one bed and a total of 5 patients are hospitalized, the bed occupancy prediction device may collect whether the bed is occupied by a patient as 5.

[0048] In step (220), a bed occupancy prediction device according to one embodiment can generate ward data based on time-dependent patient occupancy information in a ward containing each bed from the collected bed information data. Table 1 below illustrates an example of ward data generated by the bed occupancy prediction device.

[0049] Ward Data: Ward abbreviation, year of ward data collection, month of collection, day of collection, day of the week, weekend, public holiday, ward staffing, ward capacity, ward occupancy rate

[0050] In Table 1, ward abbreviations represent data used to identify the ward containing the corresponding bed in the collected bed information. For example, a ward abbreviation can be numeric data combining the "building" and "floor" within the hospital to which the corresponding bed belongs. For example, if the hospital includes buildings such as the West Wing, East Wing, and New Wing, the ward abbreviation for Ward 71, located on the 7th floor of the West Wing, can represent the numeric data "71," the ward abbreviation for Ward 73, located on the 7th floor of the East Wing, can represent the numeric data "73," and the ward abbreviation for Ward 75, located on the 7th floor of the New Wing, can represent the numeric data "75." Among ward abbreviations, English abbreviations can be excluded as data used to identify wards with anomalous admission and discharge frequencies, such as intensive care units. For example, English abbreviations such as Surgery (or GS), Orthopedics (or PS), and Neurosurgery (or NS) can be excluded. The bed occupancy prediction device can distinguish which ward the collected bed information data belongs to through the ward abbreviation. In Table 1, the ward data collection year, collection month, collection day, and collection day indicate information corresponding to the time the ward data was collected. For example, if the ward data was collected on Thursday, November 9, 2023, the bed occupancy prediction device can generate ward data by distinguishing the ward data collection year as 2023, the collection month as November, the collection day as 09, and the collection day as Thursday. Because the ward data includes data indicating the collection year, collection month, collection day, and collection day, the ward data can represent dynamic data containing time-series information. Specific descriptions of the ward data in Table 1, including whether it is a weekend, whether it is a public holiday, ward staffing, ward capacity, and ward occupancy ratio, are provided in detail in Figure 3 below.

[0051] In step (230), a bed occupancy prediction device according to one embodiment may generate room data based on at least one of time-dependent patient occupancy information or time-independent room identification information for each room containing each bed from the collected bed information data. Table 2 below illustrates examples of room data generated by the bed occupancy prediction device.

[0052] Hospital room data, patient occupancy information, hospital room abbreviation, year of hospital room data collection, month of collection, day of collection, day of the week, weekend, holiday, hospital room capacity, hospital room occupancy rate, hospital room identification information, hospital room class code, nuclear medicine ward, sterile room, isolation ward, EEG examination room, observation room, kidney transplant ward, liver transplant ward, SUBICU, special room, small single room, short-term ward, double room in psychiatry, open ward in psychiatry

[0053] In Table 2, descriptions of ward abbreviations related to patient occupancy information, year of ward data collection, month of collection, day of collection, day of collection, weekend, holiday, ward capacity, ward capacity, and ward occupancy ratio are omitted as they overlap with the ward data described in Table 1.

[0054] In Table 2, time-independent room identification information can represent static data. In other words, room identification information can represent data that remains constant even when bed information data is collected periodically over time. Room identification information can represent information that indicates the unique characteristics of the room containing the bed for which bed information data is collected. For example, if the disease characteristics of an inpatient are not anomalous, it can represent information about the room in which the patient was admitted. In other words, room identification information in Table 2 can exclude information about rooms with frequent admissions and discharges over time. For example, information such as whether the room is a delivery room, an intensive care unit (ICU), or a nuclear medicine treatment room (NMTR) can be excluded from room identification information because the patient occupancy of the room frequently changes over time. In other words, the bed occupancy prediction device can generate room data based on room identification information by considering the characteristics of rooms used in non-emergency situations.

[0055] In step (240), a bed occupancy prediction device according to one embodiment may train a machine learning-based occupancy prediction model using training data generated based on dividing at least one of time-dependent information and ward data from ward data into predetermined time periods. The bed occupancy prediction device may generate training data by dividing at least one of patient occupancy information and ward data in Table 2 into predetermined time periods from ward data. For example, the bed occupancy prediction device may generate training data including ward data for every three days or training data including ward data for every seven days through a window that divides the patient occupancy information in Table 2 into data units collected for three days or seven days. As another example, the bed occupancy prediction device may generate training data including ward data for every seven days or training data including ward data for every 30 days through a window that divides ward data into data units collected for seven days or thirty days. The bed occupancy prediction device can train a machine learning-based occupancy prediction model using training data generated based on windows that divide the collected data into 3-day, 7-day, or 30-day units.

[0056] In step (250), a bed occupancy prediction device according to one embodiment can output occupancy information of at least one ward or room by applying a trained machine learning-based occupancy prediction model to any input bed information data.

[0057]

[0058] FIG. 3 is a block diagram of a bed occupancy prediction operation of a bed occupancy prediction device according to one embodiment.

[0059] According to one embodiment, a bed occupancy prediction device may generate ward data and room data based on periodically collected bed information data. For example, the bed occupancy prediction device may generate ward dynamic data (310), room dynamic data (311), and room static data (312). For example, the ward dynamic data (310) may include accumulated bed information data for beds included in a target ward among bed information data collected at preset time intervals. The bed occupancy prediction device may determine ward resource information and ward occupancy ratio for each ward including each bed from a result of combining point-in-time data that distinguishes the time point of collection of the bed information data and the collected bed information data, and the point-in-time data may include information indicating whether the collection time point falls on at least one of a weekend or a public holiday. For example, the bed occupancy prediction device may store a value of 0 if the time point at which the bed information data is collected does not fall on a public holiday, a value of 1 if the time point at which the bed information data is collected falls on a public holiday, and may store column data indicating whether the time point is a public holiday. The bed occupancy prediction policy can generate ward dynamic data (310) by combining column data indicating whether it is a public holiday with collected bed information data.

[0060] For reference, the proportion of patients admitted to hospitals on weekends or public holidays may differ from the proportion of patients admitted to hospitals on weekdays. For example, a hospital ward occupancy rate on a weekend or public holiday may be approximately 80%, while on a non-weekend or public holiday, the occupancy rate may be approximately 89%, representing a difference of approximately 9%. For another example, a hospital bed occupancy rate on a weekend or public holiday may be approximately 79%, while on a non-weekend or public holiday, the occupancy rate may be approximately 88%, representing a difference of approximately 9%. In other words, since the difference in hospital occupancy rates between weekends and non-public holidays may be approximately 9%, distinguishing between weekends and public holidays and non-weekend or public holiday days can improve the accuracy of bed occupancy predictions. A bed occupancy prediction device can determine whether the collected bed information data falls on a weekend or public holiday by combining collected bed information data with time data indicating whether the collected bed information falls on at least one of the following: The bed occupancy prediction device can improve the accuracy of hospital bed occupancy prediction using a machine learning-based occupancy prediction model by dividing the collected bed information data into data corresponding to weekends or public holidays and data not corresponding to weekends or public holidays. The bed occupancy prediction device can derive and generate base year, base month, base week, base day, and base day of the week variables based on the date the bed information data was collected, based on the point-in-time data. For example, raw bed information data periodically collected by the bed occupancy prediction device can have a base date (e.g., 20231118) and a base time (e.g., 1530) in the form of numeric data.The bed occupancy prediction device can generate and store in a new column, based on the point-in-time data, the collected bed information data by deriving the base year (e.g., 2023), base month (e.g., 11), base week (e.g., the number corresponding to the week when a year is defined as 52 weeks), base day (e.g., 18), base time (e.g., 1500), and base day of the week (month) variables. As another example, the hospital room dynamic data (311) can include accumulated data of bed information data for beds included in a target hospital room among the bed information data collected at preset time intervals. The bed occupancy prediction device can determine the hospital room staffing information and the hospital room occupancy rate for each hospital room including each bed from the result of combining the point-in-time data that distinguishes the collection time of the bed information data and the collected bed information data, and the point-in-time data can include information indicating whether the collection time falls on at least one of a weekend or a public holiday. The description of the point-in-time data included in the ward dynamic data (311), which includes information indicating whether it falls on at least one of a weekend or a public holiday, is not repeated as it overlaps with the description of the ward dynamic data (310) above.

[0061] The hospital room static data (312) may include time-independent information related to the target hospital room to which the bed belongs. For example, as described in FIGS. 1 and 2 , the hospital room static data (312) may include identification information for the hospital room to which the target bed belongs (e.g., hospital room class code, whether it is a nuclear medicine ward, whether it is a sterile room, etc.).

[0062] According to one embodiment, a bed occupancy prediction device can generate first model input data (320) by preprocessing ward dynamic data (310). The bed occupancy prediction device can extract data indicating whether a patient can be admitted to a ward including each bed from collected bed information data. For example, if the collected bed information data includes data indicating whether a patient can be admitted to a ward including each bed or data indicating whether a patient who has been discharged is currently present, the bed occupancy prediction device can extract the data. If the extracted data indicates whether a patient can be admitted or discharged, the bed occupancy prediction device can preprocess the data into the number 0 because it means that there are no patients in the target ward, and vice versa. The bed occupancy prediction device can generate first model input data (320) including the preprocessing result of the data indicating whether a patient can be admitted (e.g., replacing 0 if a patient can be admitted, and replacing 1 if not).

[0063] A bed occupancy prediction device can generate a bed occupancy value based on ward resource information and ward capacity information for each ward. The bed occupancy prediction device can generate first model input data (320) based on the ward resource information, ward capacity information, and bed occupancy value for each ward. For example, the bed occupancy prediction device can determine data indicating ward resource information as 1 if a bed included in a target ward is occupied by a patient among the collected bed information data. In other words, the ward resource information can correspond to a value obtained by adding all data 1s indicating the presence of a patient occupying a bed included in the target ward. The ward capacity information can correspond to a value obtained by adding all the number of beds in the target ward. Therefore, the bed occupancy prediction device can generate first model input data (320) including a current ward occupancy value corresponding to a value obtained by dividing the ward resource information by the ward capacity information.

[0064] In summary, the bed occupancy prediction device can generate first model input data (320) including data indicating whether a patient can be admitted to a ward including a bed among ward dynamic data (310) preprocessed to 0 if a patient can be admitted and 1 if the patient is admitted, and a current ward occupancy value corresponding to a value obtained by dividing ward resource information by ward capacity information.

[0065] According to one embodiment, a bed occupancy prediction device can generate second model input data (321) by preprocessing hospital room dynamic data (311). The bed occupancy prediction device can extract data indicating whether a patient can enter a hospital room including each bed from collected bed information data. The method by which the bed occupancy prediction device extracts data indicating whether a patient can enter a hospital room overlaps with the process of extracting data indicating whether a patient can enter a ward described above, and thus will not be described again. The bed occupancy prediction device can generate a hospital room occupancy value based on hospital room staffing information and hospital room staffing information for each hospital room. The bed occupancy prediction device can generate second model input data (321) based on hospital room staffing information, hospital room capacity information, and hospital room occupancy value for each hospital room. The method by which the bed occupancy prediction device generates the second model input data (321) overlaps with the method of generating the first model input data (320) described above, and thus will not be described again.

[0066]

[0067] According to one embodiment, a bed occupancy prediction device may generate third model input data (322) based on hospital room data including hospital room dynamic data (311) and hospital room static data (312). For example, the third model input data (322) may include data obtained by combining the second model input data (321) and preprocessed data of the hospital room static data (312). Hereinafter, data generated by preprocessing the hospital room static data (312) will be described in detail.

[0068] The bed occupancy prediction device can separate independent bed identification information corresponding to the bed static data (312) into first identification information indicating the bed and second identification information other than the first identification information.

[0069] For reference, time-independent room identification information corresponding to room static data (312) may include information indicating a patient's non-emergency status during the period in which bed information data is periodically collected. For example, details of rooms with frequent patient admissions and discharges, such as delivery rooms and intensive care units, frequently change over time and may therefore be excluded from time-independent room identification information corresponding to room static data (312).

[0070] To explain the operation of separating the first identification information and the second identification information by the bed occupancy prediction device, it is assumed that there is room identification information for any room A in Table 3 below.

[0071] Ward A: Nuclear Medicine Ward? Sterile Room? Small Single Room? Special Room? Short-Term Ward? OOOXX

[0072] The room identification information for room A in Table 3 indicates that room A corresponds to a nuclear medicine ward, a sterile room, and a small single room. At the same time, the room meal information for room A in Table 3 indicates that room A does not correspond to a special room or a short-term ward. The bed occupancy prediction device can separate the information indicating room A in Table 3, such as whether it is a nuclear medicine ward, whether it is a sterile room, and whether it is a small single room, into first identification information, and the other information indicating whether it is a special room or a short-term ward, into second identification information.

[0073] The bed occupancy prediction device can transform the separated first and second identification information into predetermined numerical values. In the example above, the bed occupancy prediction device can transform data indicating whether the patient is in a nuclear medicine ward, a sterile room, or a small single room into a 1, and data indicating whether the patient is in a special room or a short-term ward into a 0.

[0074] The bed occupancy prediction device can generate third model input data (322) by combining a predetermined numerical value (e.g., 1 if applicable, 0 if not applicable) with data indicating whether a patient can be admitted to a hospital room, hospital room capacity information, hospital room staffing information, and hospital room occupancy value.

[0075] The bed occupancy prediction device can obtain data of a size corresponding to a predetermined period by applying windows (330, 331, 332) to the first to third model input data (320, 321, 322). For example, the bed occupancy prediction device can apply a window (330) corresponding to 7 days or 30 days to the first model input data (320) generated based on ward dynamic data (310). As another example, the bed occupancy prediction device can apply a window (331) corresponding to 3 days or 7 days to the second model input data (321) generated based on ward dynamic data (311). As yet another example, the bed occupancy prediction device can apply a window (332) corresponding to 3 days or 7 days to the third model input data (322). In other words, the bed occupancy prediction device can apply a window (330) of 7 or 30 days to the first model input data (320) based on ward data, and can apply a window (331, 332) of 3 or 7 days to the input data (321, 322) based on ward data. The period of each of the windows (330, 331, 332) may be shorter than the period of the window (330) for the ward data, because the change in patient occupancy over time in the ward is relatively greater in the ward.

[0076] The bed occupancy prediction device can output a patient occupancy prediction result for a ward by inputting data generated by applying a window (330) to the first model input data (320) into a machine learning-based occupancy prediction model (340). In addition, the bed occupancy prediction device can output a patient occupancy prediction result for a ward by inputting data generated by applying a window (331) to the second model input data (321) or data generated by applying a window (332) to the third model input data (322) into a machine learning-based occupancy prediction model (340).

[0077]

[0078] FIG. 4 illustrates a bed occupancy prediction device according to one embodiment training a machine learning-based occupancy prediction model based on different training data.

[0079] According to one embodiment, the bed occupancy prediction device can generate training data by applying a window (420) that divides ward data into predetermined time periods. For example, the bed occupancy prediction device can generate training data based on ward data divided into at least one of 7 days and 30 days. For example, the bed occupancy prediction device can generate training data by applying a window (420) (e.g., window 7) that divides ward data into 7-day units to input data (e.g., input A). A machine learning-based occupancy prediction model trained based on the training data generated by applying window 7 is hereinafter referred to as a ward 7-day model. In another example, the bed occupancy prediction device can generate training data by applying a window (420) (e.g., window 30) that divides ward data into 30-day units to input data (e.g., input A). A machine learning-based occupancy prediction model trained based on the training data generated by applying window 30 is hereinafter referred to as a ward 30-day model.

[0080] The bed occupancy prediction device can generate training data by applying windows (421, 422) that divide time-dependent information in the room data into predetermined time periods. For example, the bed occupancy prediction device can generate training data based on dividing time-dependent information in the room data into at least one of 3 days or 7 days. For example, the bed occupancy prediction device can generate training data by applying a window (421) (e.g., window 3) that divides into 3-day units to input data (e.g., input B) based only on room dynamic data among the room data. A machine learning-based occupancy prediction model trained on the training data generated by applying Window 3 to the input data based only on room dynamic data is hereinafter referred to as a room 3-day model. As another example, the bed occupancy prediction device can generate training data by applying a window (421) (e.g., window 7) that divides into 7-day units to input data (e.g., input B) based only on room dynamic data among the room data. A machine learning-based occupancy prediction model trained on training data generated by applying window 7 to input data based only on room dynamic data is referred to as a room 7 days model hereinafter. As another example, a bed occupancy prediction device can generate training data by applying a window (422) (e.g., window 3 or window 7) that divides the input data into 3-day or 7-day units (e.g., input C) based on data combining room dynamic data and room static data.The machine learning-based occupancy prediction models trained on training data generated by applying window 3 or window 7 to input data based on a combination of room dynamic data and room static data are referred to as the room static 3-day model and the room static 7-day model, respectively. Table 4 below compares the performance of machine learning-based occupancy prediction models trained on different training data.

[0081] ModelMAEMSERMSER2 ScoreWardWard 7 Days0.0570.0070.0820.582Ward 30 Days0.0620.0090.0930.458RoomRoom 3 Days0.1260.0530.2310.294Room 7 Days0.1230.0520.2270.317Room Static 3 Days0.1240.0570.2390.246Room Static 7 Days0.1230.0510.2260.320

[0082] In Table 4, MAE (mean absolute error) represents the mean absolute error, which represents the absolute difference between the model's predicted value and the actual bed occupancy rate; MSE (mean square error) is the average of the squared differences between the model's predicted value and the actual bed occupancy rate, which represents the mean squared error that is sensitive to outliers; RMSE (root mean square error) is the positive square root of MSE, which is the square root of MSE; and R2 Score is the correlation between the model's predicted value and the actual value, which shows how well the independent variables explain the dependent variable. Table 4 shows that the performance of the model trained on the training data obtained by applying window 7 to the input data generated based on the room data that combined the room dynamic data and the room static data is superior to the performance of the model trained on the room static 7 days model. In other words, Table 4 shows that among the room occupancy prediction models, the room static 7 days model has the best performance.

[0083]

[0084] FIG. 5 illustrates a machine learning-based occupancy prediction model architecture according to one embodiment.

[0085] According to one embodiment, a bed occupancy prediction device can output bed occupancy information within a hospital by applying a machine learning-based occupancy prediction model to input data based on periodically collected bed information data. The machine learning-based occupancy prediction model included in the bed occupancy prediction device can predict bed occupancy based on simultaneously receiving static data (e.g., time-independent room identification information) and dynamic data (e.g., time-dependent information among room data).

[0086] A bed occupancy prediction device can extract a first output including time-series information of the dynamic data by applying a first model (510) including an LSTM layer to dynamic data generated based on dividing at least one of time-dependent information and ward data from ward data into predetermined time periods. For example, the first model (510) may include a model that receives dynamic data (e.g., dynamic valuables, ward data, or time-dependent patient occupancy information among ward data) such as the first model input data (320) or the second model input data (321) of FIG. 3 as input, and extracts a first output including time-series information of the dynamic data. For example, the first model (510) may include a Bi-LSTM (bidirectional long short term memory) layer that processes input data in both directions.

[0087] For reference, Bi-LSTM is a type of recurrent neural network (RNN) that can be used to process sequential or sequence data. Bi-LSTM can simultaneously consider information from previous and subsequent time points regarding input data. Bi-LSTM can include two LSTM layers, with one LSTM layer processing input data from front to back, and the other LSTM layer processing input data from back to front. Bi-LSTM can merge the data processed by the two LSTM layers to generate the final output. By processing input data in both directions, Bi-LSTM can produce output that preserves the time-series information of the input data.

[0088] The first model (510) may include multiple Bi-LSTM layers. For example, the first model (510) may include Bi-LSTM layers in the first layer and the second layer.

[0089] A LeakyReLu operation may be performed in the Bi-LSTM layer of the first model (510). For reference, the LeakyReLu operation may represent an operation that multiplies an input by a very small value that is not 0 and outputs it, even if the input is negative.

[0090] The first model (510) may include an attention layer. For example, the first model (510) may include an attention layer connected to a Bi-LSTM layer. The attention layer included in the first model (510) can generate data focused on values ​​among the Bi-LSTM layer's outputs that have a significant impact on the occupancy prediction results by assigning high weights to those values.

[0091] The bed occupancy prediction device can extract a second output including features of the static data by applying a second model (520) including a dense layer to static data generated based on time-independent room identification information from the room data. For example, the second model (520) can include a model that receives static data (e.g., static valuables, room identification information from the room data, etc.) as input and extracts a second output including features of the static data. For example, the second model (520) can receive static data based on the room static data (312) among the third model input data (322) of FIG. 3 as input and generate a second output including features of the room static data.

[0092] The second model (520) may include a dense layer. A dropout operation may be performed in the dense layer included in the second model. For reference, the dropout operation may refer to an operation that removes neurons with a probability between 0 and 1 in interconnected layers. For example, assuming a dropout ratio of 0.5 and the presence of four neurons (or nodes), the dropout operation may refer to an operation that removes each of the four neurons with a probability of 0.5. Therefore, the dropout operation can prevent overfitting, which may occur due to excessive learning of features of only specific variables.

[0093] The bed occupancy prediction device can output occupancy information of at least one ward or room based on the concatenation of the first output and the second output. The bed occupancy prediction device can concatenate the first output and the second output. By concatenating the first output and the second output, the bed occupancy prediction device can preserve not only the time-series characteristics of dynamic data but also the characteristics of static data. The bed occupancy prediction device can output bed occupancy information having a value between 0 and 1 by inputting the data concatenated with the first output and the second output into a dense layer (540).

[0094]

[0095] FIGS. 6A and 6B illustrate user interface displays for hospital bed occupancy prediction results predicted by a bed occupancy prediction device according to one embodiment.

[0096] Figure 6a shows a user interface that displays the current status and outlook for bed occupancy in a hospital ward.

[0097] For example, a bed occupancy prediction device can output current occupancy information (610a) and expected occupancy information (620a) of all beds in a ward based on ward data. The bed occupancy prediction device can output the current occupancy information (610a) and expected occupancy information (620a) in a graph form. For example, the bed occupancy prediction device can output the ratio of currently occupied beds (e.g., in use 86%) and the ratio of available beds (e.g., empty 11%) for the current occupancy information (610a). The bed occupancy prediction device can output the number of currently occupied beds (e.g., currently used beds 458) and the number of available beds (e.g., available beds) for the current occupancy information (610a).

[0098] For example, a bed occupancy prediction device can graph actual and predicted occupancy information for a ward together with predicted occupancy information (620a). A user can specify a date for which they wish to check the occupancy information in the predicted occupancy information (620a). For example, the user can click on the date for which they wish to check the bed occupancy information in the predicted occupancy information (620a) graph. In response to the user's click on the graph, the bed occupancy prediction device can display the predicted occupancy rate of beds in the ward for the selected date (630a).

[0099] Figure 6b shows a user interface that displays the current status and outlook of bed occupancy in a hospital ward.

[0100] For example, the bed occupancy prediction device can output current occupancy information and expected occupancy information for beds in individual wards to the user interface (610b) based on ward data. For example, the bed occupancy prediction device can output current occupancy information of 88.9% and expected occupancy information of 79.1% for the ward corresponding to MICU1_01. The bed occupancy prediction device can display the number of occupied beds (621b), the number of reserved beds (622b), the number of remaining beds (623b), and the number of beds being cleaned (624b) in individual wards to the user interface (610b).

[0101]

[0102] The embodiments described above may be implemented using hardware components, software components, and / or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may be implemented using a general-purpose computer or a special-purpose computer, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing instructions and responding to them. The processing device may execute an operating system (OS) and software applications running on the operating system. Furthermore, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing device is sometimes described as being used alone; however, one of ordinary skill in the art will recognize that the processing device may include multiple processing elements and / or multiple types of processing elements. For example, a processing unit may include multiple processors, or a processor and a controller. Other processing configurations, such as parallel processors, are also possible.

[0103] Software may include computer programs, codes, instructions, or a combination of one or more of these, which may configure a processing device to perform a desired operation or may independently or collectively command the processing device. The software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, or computer storage medium or device for interpretation by the processing device or for providing instructions or data to the processing device. The software may also be distributed over networked computer systems and stored or executed in a distributed manner. The software and data may be stored on a computer-readable recording medium.

[0104] The method according to the embodiment may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program commands, data files, data structures, etc., alone or in combination, and the program commands recorded on the medium may be those specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program commands such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

[0105] The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

[0106] In this document, each of the phrases "A or B", "at least one of A and B", "at least one of A or B", "A, B, or C", "at least one of A, B, and C", and "at least one of A, B, or C" may include any one of the items listed together in that phrase, or all possible combinations thereof.

[0107] Although the embodiments described above have been described with limited drawings, those skilled in the art will appreciate that various technical modifications and variations can be applied based on the described embodiments. For example, appropriate results can still be achieved even if the described techniques are performed in a different order than described, and / or components of the described systems, structures, devices, circuits, etc. are combined or combined in a different manner than described, or are replaced or substituted with other components or equivalents.

[0108] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims described below.

Claims

1. In a method for predicting bed occupancy performed by a processor, A step of periodically collecting bed information data including whether each bed is occupied by a patient and room identification information indicating the corresponding bed; A step of generating ward data based on time-dependent patient occupancy information in a ward including each bed from the collected ward information data; A step of generating hospital room data based on at least one of time-dependent patient occupancy information or time-independent hospital room identification information in a hospital room including each bed from the collected hospital bed information data; A step of training a machine learning-based occupancy prediction model using training data generated based on time-dependent information from the above ward data and dividing at least one of the above ward data into predetermined period units; and A step of applying the above trained machine learning-based occupancy prediction model to the input hospital bed information data, thereby outputting occupancy information of at least one of a ward or a hospital room. Including A method for predicting hospital bed occupancy.

2. In paragraph 1, The steps for generating the above ward data are: A step for determining ward resource information and ward occupancy rate for each ward including each bed from the results of combining the time point data that distinguishes the time point of collection of the above ward information data and the collected ward information data. Including, The above point data is, Including information indicating whether the said collection time falls on at least one of a weekend or public holiday; A method for predicting hospital bed occupancy.

3. In paragraph 2, The steps for generating the above ward data are: A step of extracting data indicating whether or not a patient can be admitted to a ward containing each bed from the collected bed information data; and Step of generating a bed occupancy value based on ward resource information and ward capacity information for each ward above Including, A method for predicting hospital bed occupancy.

4. In paragraph 1, The steps for generating the above ward data are: A step for determining the hospital room occupancy rate and hospital room staffing information for each hospital room including each bed from the results of combining the collected hospital bed information data and the time point data that distinguishes the time point of collection of the above hospital bed information data. Including, The above point data is, Including information indicating whether the said collection time falls on at least one of a weekend or public holiday; A method for predicting hospital bed occupancy.

5. In paragraph 4, The steps for generating the above ward data are: A step of extracting data indicating whether a patient can enter a ward containing each bed from the collected bed information data; and Step for generating a hospital room occupancy rate value based on the hospital room resource information and hospital room capacity information for each hospital room Including A method for predicting hospital bed occupancy.

6. In paragraph 5, The steps for generating the above ward data are A step of dividing the independent ward identification information at the above time into first identification information indicating the ward and second identification information other than the first identification information; A step of transforming the separated first identification information and the second identification information into predetermined numerical values, respectively; and A step of combining the above predetermined numerical value and data indicating whether the patient can enter the above ward, the ward capacity information, the ward resource information, and the ward occupancy rate value. Including Independent ward identification information at the above time During the period in which the above hospital bed information data is periodically collected, information indicating the patient's non-emergency is included. A method for predicting hospital bed occupancy.

7. In paragraph 1, The step of training the above machine learning-based occupancy prediction model is: A step of generating training data based on dividing the above ward data into at least one period unit of 7 days or 30 days. Including A method for predicting hospital bed occupancy.

8. In paragraph 1, The step of training the above machine learning-based occupancy prediction model is: A step of generating training data based on dividing time-dependent information in the above ward data into at least one period unit of 3 days or 7 days. Including A method for predicting hospital bed occupancy.

9. In paragraph 1, The step of outputting the occupancy information of at least one of the above wards or rooms is: A step of extracting a first output including time-series information of the dynamic data by applying a first model including an LSTM layer (long short term memory layer) to dynamic data generated based on dividing at least one of the time-dependent information in the ward data and the ward data into predetermined time units; A step of extracting a second output including features of the static data by applying a second model including a dense layer to static data generated based on time-independent ward identification information from the ward data; and A step of outputting occupancy information of at least one of the wards or rooms based on a concatenation of the first output and the second output. Including A method for predicting hospital bed occupancy.

10. A computer program stored on a computer-readable recording medium for executing the method of any one of claims 1 to 9 in combination with hardware.

11. In the bed occupancy prediction device, A processor that periodically collects bed information data including whether a bed is occupied by a patient and bed identification information indicating the corresponding bed, generates ward data based on time-dependent patient occupancy information in a ward including each bed from the collected bed information data, generates hospital room data based on at least one of time-dependent patient occupancy information or time-independent hospital room identification information in a ward including each bed from the collected bed information data, trains a machine learning-based occupancy prediction model using training data generated based on dividing at least one of the time-dependent information and the ward data in the ward data into predetermined period units, and applies the trained machine learning-based occupancy prediction model to input bed information data, thereby outputting occupancy information of at least one of a ward or a hospital room. Including Bed occupancy prediction device.

12. In paragraph 11, The above processor, The ward data is generated by determining the ward resource information and ward occupancy rate for each ward including each bed from the result of combining the point-in-time data that distinguishes the point in time when the above ward information data is collected and the collected ward information data. The above point data is, Contains information indicating whether the above collection time falls on at least one of a weekend or public holiday. Bed occupancy prediction device.

13. In paragraph 12, The above processor, From the collected bed information data, data indicating whether a patient can be admitted to a ward including each bed is extracted, and a bed occupancy value is generated based on ward resource information and ward capacity information for each ward, thereby generating the ward data. Bed occupancy prediction device.

14. In paragraph 11, The above processor, The hospital room data is generated by determining the hospital room occupancy rate and hospital room staffing information for each hospital room including each bed from the results of combining the time point data that distinguishes the time point of collection of the above hospital bed information data and the collected hospital bed information data. The above point data is, Including information indicating whether the said collection time falls on at least one of a weekend or public holiday; Bed occupancy prediction device.

15. In paragraph 14, The above processor, From the collected bed information data, data indicating whether a patient can enter a room including each bed is extracted, and a bed occupancy rate value is generated based on the bed capacity information and bed capacity information for each bed, thereby generating the bed data. Bed occupancy prediction device.

16. In paragraph 15, The above processor, The independent hospital room identification information at the above time is separated into first identification information indicating the corresponding hospital room and second identification information other than the first identification information, the separated first identification information and the second identification information are each transformed into a predetermined numerical value, and the predetermined numerical value is combined with data indicating whether the hospital room can accommodate a patient, the hospital room capacity information, the hospital room resource information, and the hospital room occupancy value, thereby generating the hospital room data. Independent ward identification information at the above time is, During the period in which the above hospital bed information data is periodically collected, information indicating the patient's non-emergency is included. Bed occupancy prediction device.

17. In paragraph 11, The above processor, By generating training data based on dividing the above ward data into at least one period unit of 7 days or 30 days, Training the above machine learning-based occupancy prediction model Bed occupancy prediction device.

18. In paragraph 11, The above processor, By generating training data based on dividing the time-dependent information in the above ward data into at least one period unit of 3 days or 7 days, Training the above machine learning-based occupancy prediction model Bed occupancy prediction device.

19. In paragraph 11, The above processor, By applying a first model including an LSTM layer (long short term memory layer) to dynamic data generated based on dividing at least one of the time-dependent information and the ward data into predetermined time units in the ward data, a first output including time-series information of the dynamic data is extracted, and a second model including a dense layer is applied to static data generated based on time-independent ward identification information in the ward data, a second output including features of the static data is extracted, and based on a concatenation of the first output and the second output, occupancy information of at least one of the ward or ward is output. Bed occupancy prediction device.

Images