Computer, information processing system, and information processing method

Abstract

This computer comprises a processor (control unit 100) and a memory (storage unit 110), wherein the processor executes a program stored in the memory to acquire a plurality of pieces of assessment target information indicating each of a plurality of types of assessment targets pertaining to a patient, and unify the pieces of assessment target information about the patient into a reference format preset for each type. The reference format includes a static format not including time and a time series format including time, and the time series format includes a point time series format indicating an assessment target at a prescribed time point and an interval time series format indicating an assessment target in a prescribed interval.

Description

Computer, Information Processing System, and Information Processing Method

[0001] The present disclosure relates to a computer, an information processing system, and an information processing method.

[0002] Conventionally, learning models obtained by machine learning have been used in various scenarios, and they are also used in the medical field. Japanese Patent Laid-Open Publication No. 2022-183393 (JP2022-183393A) discloses a prediction determination model that can accurately predict CRT non-responders. This prediction determination model is obtained by machine learning using data of patients who may become non-responders as teacher data.

[0003] There are cases where it is desired to use data obtained in medicine, such as learning data in machine learning. In such cases, it is desirable that the data formats are unified to some extent. However, for example, in drug administration information, in the information system of a certain medical institution, the amount of drug solution administered every hour is recorded as data, while in the information system of other medical institutions, only the start time and end time of drug administration and the administration rate are recorded as data. Due to differences in the way data is held in the information systems of each medical institution, etc., different forms of data may be recorded for the same treatment. When data of different formats is used as learning data, there is a problem that it is difficult to generate a desired prediction model.

[0004] The present disclosure has been made in consideration of such points, and aims to convert the input information into a format suitable for machine learning.

[0005] The computer of this disclosure is a computer comprising a processor and memory, wherein the processor, by executing a program stored in the memory, acquires a plurality of evaluation target information indicating each of a plurality of types of evaluation targets relating to a patient, unifies the evaluation target information of the patient into a predetermined standard format for each type, the standard format includes a static format that does not include time and a time series format that includes time, and the time series format includes a point time series format indicating an evaluation target at a predetermined point in time and an interval time series format indicating an evaluation target in a predetermined interval.

[0006] The information processing system disclosed herein is an information processing system comprising an acquisition unit and a unification unit, wherein the acquisition unit acquires a plurality of evaluation target information indicating each of a plurality of types of evaluation targets relating to a patient, and the unification unit unifies the evaluation target information of the patient into a standard format predetermined for each type, the standard format includes a static format and a time series format including time, and the time series format includes a point time series format indicating an evaluation target at a predetermined point in time and an interval time series format indicating an evaluation target in a predetermined interval.

[0007] The information processing method of the present disclosure is an information processing method performed by a computer having a processor, wherein the processor acquires a plurality of evaluation target information indicating each of a plurality of types of evaluation targets relating to a patient, and the processor unifies the evaluation target information of the patient into a predetermined standard format for each type, wherein the standard format includes a static format and a time series format including time, and the time series format includes a point time series format indicating an evaluation target at a predetermined time and an interval time series format indicating an evaluation target in a predetermined interval.

[0008] This is an overall diagram of the information processing system. This is a flowchart of the model management process. This is a diagram showing an example of the data structure of medical data. This is a flowchart showing detailed processing in the medical terminology standardization process. This is a diagram showing an example of how to write a string of characters. This is a diagram showing an example of the data structure of treatment information related to intravenous administration at Hospital A. This is a diagram showing an example of the data structure of treatment information related to intravenous administration at Hospital B. This is a diagram showing an example of the data structure of treatment information related to intravenous administration recorded in interval time series format. This is a diagram showing an example of the data structure of treatment information related to the use of a ventilator at Hospital A. This is a diagram showing an example of the data structure of treatment information related to the use of a ventilator at Hospital B. This is a diagram showing an example of the data structure of treatment information related to the use of a ventilator recorded in interval time series format.

[0009] The embodiment will now be described with reference to the drawings. Figure 1 is an overall configuration diagram of the information processing system 1 according to this embodiment. The information processing system 1 according to this embodiment acquires medical data generated at each of multiple medical institutions and generates a sudden change prediction model that predicts sudden changes in a patient's condition using the medical data as training data. Furthermore, the information processing system 1 according to this embodiment performs sudden changes in a patient's condition by using the sudden change prediction model generated in this way.

[0010] "Sudden deterioration" refers to a predetermined symptom that necessitates new treatment. Examples of "sudden deterioration" include sudden deterioration requiring admission to the intensive care unit, sudden deterioration in arterial oxygen saturation requiring oxygen administration or mechanical ventilation, sudden deterioration in blood pressure or pulse requiring vasopressors, massive fluid resuscitation or blood transfusion, the development of disseminated intravascular coagulation requiring treatment, and impending or reversible cardiac arrest requiring vasopressors, massive fluid resuscitation, blood transfusion or aortic clamping. Furthermore, it includes the initiation of cardiopulmonary resuscitation or death in hospital as a result of any of these. Typically, "sudden deterioration" is a sudden deterioration in arterial oxygen saturation requiring oxygen administration or mechanical ventilation, sudden deterioration in blood pressure or pulse requiring vasopressors, massive fluid resuscitation or blood transfusion, the development of disseminated intravascular coagulation requiring treatment, and impending or reversible cardiac arrest requiring vasopressors, massive fluid resuscitation, blood transfusion or aortic clamping.

[0011] The information processing system 1 generates a sudden change prediction model and uses it to predict sudden changes. However, as an alternative, the information processing system 1 may predict not only sudden changes but also any predetermined state of the patient. Specifically, the information processing system 1 may generate a prediction model to predict the patient's state and use this prediction model to predict the state of the target patient. Examples of such states include decreased blood pressure and decreased oxygen saturation.

[0012] The information processing system 1 comprises a model management device 10, a user terminal 20, and multiple hospital servers 30. The model management device 10 generates a sudden change prediction model and performs sudden change prediction using the sudden change prediction model. The user terminal 20 is an information processing device used by users of the model management device 10. The two hospital servers 30 are located in different hospitals (Hospital A and Hospital B). In this embodiment, for the sake of explanation, only two hospital servers 30 are shown, but the information processing system 1 may have three or more hospital servers 30. In this case, the three or more hospital servers 30 are each located in a different hospital.

[0013] The model management device 10, user terminals 20, and multiple hospital servers 30 are connected via network N.

[0014] The model management device 10 is configured, for example, by a computer and mainly comprises a control unit 100 (processor), a storage unit 110 (memory), and a communication unit 120.

[0015] The control unit 100 includes a processor such as a CPU (Central Processing Unit) and controls the operation of the model management device 10. The communication unit 120 includes a communication interface that communicates with external devices wirelessly or via wired connection. The control unit 100 transmits and receives data between the user terminal 20 and the hospital server 30 via the communication unit 120.

[0016] The storage unit 110 includes, for example, an HDD (Hard Disk Drive), RAM (Random Access Memory), ROM (Read Only Memory), and SSD (Solid State Drive). Furthermore, the storage unit 110 is not limited to one built into the model management device 10, but may be a storage medium that can be detachably attached to the model management device 10 (for example, a USB memory). The storage unit 110 stores programs executed by the control unit 100, sudden change prediction models, and various other data.

[0017] The storage unit 110 of this embodiment includes a learning data DB 111 and a standard medical term DB 112. The learning data DB 111 stores learning data corresponding to medical data acquired from the hospital server 30. The standard medical term DB 112 stores medical terms that have been pre-set as standard medical terms. Medical terms are terms used in the medical field, such as disease names, treatment names, and equipment names. For example, for disease names, Japanese disease names corresponding to ICD-10 are stored as standard medical terms (standard disease names). Note that the standard medical term DB 112 only needs to contain pre-registered medical terms and is not limited to Japanese disease names corresponding to ICD-10.

[0018] The user terminal 20 is composed of a computer or the like and mainly comprises a communication unit 200, a control unit 210, a display unit 220, and an operation unit 230. The communication unit 200 is similar to the communication unit 120 of the model management device 10 and includes a communication interface for communicating with external devices wirelessly or via wired connection.

[0019] The control unit 210 is similar to the control unit 100 of the model management device 10, includes a processor, and controls the operation of the user terminal 20. The display unit 220 is, for example, a monitor, and displays various screens. The display unit 220 displays, for example, medical data, the results of sudden change predictions, etc. The operation unit 230 is, for example, a keyboard, and can give various commands to the control unit 210.

[0020] The hospital server 30 is composed of computers and the like. The configuration of the hospital server 30 is the same as that of the model management device 10, and mainly consists of a control unit, a storage unit, and a communication unit.

[0021] Next, the configuration of the control unit 100 of the model management device 10 will be described. The control unit 100 functions as an acquisition unit 101, a unification unit 102, a learning unit 103, and a prediction unit 104 by executing a program stored in the storage unit 110. In the following, the processes described as being executed by the acquisition unit 101, the unification unit 102, the learning unit 103, and the prediction unit 104 are processes performed by the control unit 100 by executing a program.

[0022] The acquisition unit 101 acquires medical data from each of the multiple hospital servers 30 via the communication unit 120. The unification unit 102 unifies the terminology included in the medical data to the standard medical terminology registered in the standard medical terminology DB 112. The unification unit 102 also unifies the format of each data included in the medical data. The learning unit 103 uses the medical data, which has been unified in terms and format by the unification unit 102, as training data to train the sudden change prediction model. The prediction unit 104 performs sudden change prediction by inputting the patient's medical data into the sudden change prediction model obtained through training by the learning unit 103. The processing of each functional unit will be described in detail later.

[0023] Figure 2 is a flowchart showing the model management process by the model management device 10. In the model management process, first, the acquisition unit 101 acquires medical data from each of the multiple hospital servers 30 (step S100). The medical data includes, for example, strings entered by user operations in the electronic medical record.

[0024] Figure 3 shows an example of the data structure of medical data. Medical data includes patient information, treatment information, and measurement results. Patient information includes basic patient information such as name, age, and blood type. Treatment information includes information about treatment, such as the use of a ventilator, oral administration, injection administration, and intravenous administration. Measurement results include measurement results, i.e., the patient's biological information, such as electrocardiogram waveforms, arterial pressure waveforms, and arterial blood oxygen saturation waveforms. Furthermore, the measurement results also include information on whether or not there was a sudden change in the patient's condition.

[0025] In Figure 2, after the processing in step S100, the unification unit 102 unifies the medical terms included in the medical data to the standard medical terms registered in the standard medical terminology DB 112 (step S102). The medical data includes medical terms used in medicine, such as disease names, drug names, procedure names, and equipment names. However, for example, the term "atlas fracture" may be written as "atlas fracture" or as "C1 fracture." Similarly, for "radial fracture" and "ulnar fracture," both may be written, or they may be combined into a single term such as "radial-ulnar fracture."

[0026] For medical data to be used as training data for a sudden change prediction model, it is preferable that medical terms with the same meaning be registered under the same term. Therefore, in this embodiment, the unification unit 102 performs a process to unify each medical term included in the medical data into standard medical terms.

[0027] For example, "atlas fracture" is registered as a standard medical term in the Standard Medical Terminology Database 112. Therefore, "atlas fracture" included in medical data is stored as "atlas fracture" in the training data database 111. On the other hand, "C1 fracture" included in medical data is converted (unified) to its corresponding standard medical term, "atlas fracture."

[0028] After the processing of step S102, the unification unit 102 unifies the format of the numerical data included in the medical data into a predetermined standard format (step S104). This processing will be described in detail later. Next, the unification unit 102 stores, as learning data, the medical data in which the medical terms have been unified and unified into the standard format by the above processing in the learning data DB111 (step S106).

[0029] Next, the learning unit 103 generates a sudden change prediction model by using the learning data stored in the learning data DB111 (step S108). Specifically, the learning unit 103 generates a sudden change prediction model by supervised machine learning using the learning data stored in the learning data DB111. Next, the prediction unit 104 performs a sudden change prediction using the sudden change prediction model based on the medical data of the patient to be predicted (step S110). Thus, the model management process is completed.

[0030] Note that the execution order of the medical term unification process in step S102 and the medical data format unification process in step S104 is not limited to the embodiment. As another example, step S102 and step S104 may be performed in parallel, or the process of step S102 may be performed after the process of step S104.

[0031] FIG. 4 is a flowchart showing detailed processing in the medical term unification process (step S102). In the medical term unification process, a process of unifying medical terms is performed. Hereinafter, the medical term unification process will be described by taking a disease name as an example.

[0032] In the medical term unification process (step S102), the unification unit 102 first acquires the character string included in the medical data acquired in step S100 as a target character string to be processed, and as a preprocessing process, unifies the notation fluctuations of Chinese characters and symbols (step S200). In the preprocessing process, the unification such as from "颈" to "頚" is performed.

[0033] Next, the unification unit 102 checks whether the target string after preprocessing is an exact match with a standard medical term stored in the standard medical term DB 112 (step S202). If the target string is an exact match with a standard medical term (Y in step S202), the unification unit 102 recognizes the string as a standard medical term (step S204) and completes the process.

[0034] On the other hand, if the target string does not exactly match a standard medical term (N in step S202), the unification unit 102 performs a line splitting process (step S206). In the line splitting process (step S206), the unification unit 102 splits the target string into multiple target strings according to the line splitting rule. The line splitting rule splits a target string written across multiple lines into multiple target strings. According to the line splitting rule, for example, as shown in Figure 5, the unification unit 102 splits a string displayed across two lines into two strings, each of which becomes a target string.

[0035] In the example in Figure 5, the first line contains the text "radial fracture" and the second line contains the text "ulnar fracture". In this case, "radial fracture ulnar fracture" is the target string, but in the line splitting process (step S206), the target string is split into two: "radial fracture" and "ulnar fracture". Similarly, if the target string has three or more lines, it is split into three or more target strings. If this process results in a change, that is, if the target string is split into two or more target strings (Y in step S208), the unification unit 102 proceeds to step S202 and performs the processing from step S202 onwards for each target string.

[0036] Furthermore, if there are no changes (N in step S208), the unification unit 102 proceeds to step S210. In step S210, the unification unit 102 performs numerical expansion processing according to the numerical expansion rules. The numerical expansion rules rewrite notations that include multiple contents, such as notations indicating a numerical range, into notations for each individual content. For example, an atlas fracture may be written as a C1 fracture. In addition, there may be notations that group multiple bones together, such as "C1-C3 fractures". In response to this, the unification unit 102 expands the notation grouped as "C1-C3 fractures" into multiple notations corresponding to multiple disease names, such as "C1, 2, 3 fractures," according to the numerical expansion rules. Similarly, "TH2-TH7 fractures" is expanded to "TH2, 3, 4, 5, 6, 7 fractures."

[0037] If a change occurs during the numerical expansion process, i.e., if numerical expansion is performed (Y in step S212), the unification unit 102 proceeds to step S202 and performs the subsequent processing using the numerically expanded string as the target string. If there is no change (N in step S212), the unification unit 102 proceeds to step S214. In step S214, the unification unit 102 performs string conversion processing according to the string conversion rules.

[0038] The string conversion rules convert predetermined strings into standard medical terms that are pre-associated with these strings. For example, if any of the characters "vertebral arch," "spinous process," "vertebral body," or "fracture" are detected after C7, the string conversion rules will convert them to "7th cervical vertebral arch," "7th cervical vertebral spinous process," "7th cervical vertebral body," and "7th cervical vertebral fracture," respectively. In addition, the string conversion process removes predetermined characters from within parentheses. Furthermore, expressions that modify a disease name after it, such as "second-degree burn," are changed to modify it from the beginning, such as "second-degree burn." If a change occurs in the string conversion process (step S214), i.e., if a string conversion is performed (Y in step S216), the unification unit 102 proceeds to step S202 and performs the subsequent processing using the converted string as the target string.

[0039] Also, when there is no change (N in step S216), the unifying unit 102 advances the process to step S218. In step S218, the unifying unit 102 performs a partial match process according to the partial match rule. The partial match rule detects the presence or absence of a partial match by comparing a part of the target character string with the standard medical terms in the standard medical term DB 112. In the partial match process (step S218), the unifying unit 102 further, according to the partial match rule, when adding a word at the end of a disease name such as "disease", "injury", "symptom", "syndrome", etc. to the target character string, if it matches the standard medical term, the target character string is regarded as a partial match.

[0040] And when there is a partial match (Y in step S218), the unifying unit 102 determines that the matched character string is recognized as a standard medical term (step S220) and completes the process. For example, when only "depression" is described instead of "depressive disorder", it is converted to "depressive disorder" and recognized as a standard medical term. Further, when "depression" is described, in the previous preprocessing (step S200), "depression" is converted to "depression", and in step S220, it is converted to "depressive disorder" and then recognized as a standard medical term.

[0041] Also, in the partial match process (step S218), when the target character string contains predetermined characters such as "right", "left", "both sides", "multiple", "worsening", "acute", etc. at the beginning, the unifying unit 102 deletes them, and if the character string after deletion matches the standard medical term, the target character string is regarded as a partial match. For example, when "right femoral fracture" is the target character string, "right" is deleted and it becomes "femoral fracture". Since "femoral fracture" matches the standard medical term, it is recognized as a standard medical term. Also, when "bilateral rib fracture" is the target character string, "both sides" is deleted and it becomes "rib fracture". Since "rib fracture" matches the standard medical term, it is recognized as a standard medical term. Also, for example, when "acute exacerbation of pulmonary edema" is the target character string, "acute" and "exacerbation" are deleted and it becomes "pulmonary edema". And since "pulmonary edema" matches the standard medical term, it is recognized as a standard medical term.

[0042] On the other hand, in some cases, a string containing the character "multiple", such as "multiple myeloma", is registered as a standard medical term. For such a string, in the process of checking whether it is an exact match (step S202) immediately after the preprocessing (step S200), it is recognized as matching the standard medical term. Therefore, it is possible to prevent the necessary "multiple" from being deleted by the partial match process.

[0043] Also, the partial match process is executed after the string conversion process. For example, in the string conversion process, "second-degree burn" is converted to "second-degree burns" (preposing the post-modifier). If the partial match process is performed before this process, since "burn" matches the standard medical term, "burn" is determined to be a partial match, and the information of "second-degree" is lost. On the other hand, by executing the string conversion process before the partial match process, such loss of information can be prevented.

[0044] Furthermore, in the partial match process, a process of replacing a certain word with a word of a broader concept and then determining again whether there is a partial match is performed. For example, the "ascending colon" is part of the "colon", and the "colon" is part of the "large intestine". That is, the "colon" includes the "ascending colon", and the "large intestine" includes the "colon", showing an inclusion relationship. In contrast, when the target string is "ascending colon cancer" and "ascending colon cancer" is not registered as a standard medical term, it is aggregated into "colorectal cancer". In such a case, the partial match process is performed on the target string "ascending colon cancer". Since there is no match, "ascending colon" is converted to "colon", a larger concept that includes "ascending colon", and the partial match process is performed again with "colon cancer". If there is still no match in this case, the partial match process is performed again with "colorectal cancer". And if a match is determined.

[0045] For example, in the string conversion process, if "ascending colon" is uniformly converted to "large intestine", all standard medical terms containing "ascending colon" will also be converted to "large intestine", which is not preferable. Therefore, in the present embodiment, in this way, in the partial match process, conversion to medical terms that are sequentially included is performed. Thereby, accurate conversion to standard medical terms is performed.

[0046] In the process of partial match, when no partial match is found (N in step S218), the unifier 102 advances the process to step S222. In step S222, the unifier 102 performs an expansion process. In the expansion process (step S222), the unifier 102 divides the collectively described disease names into respective disease names.

[0047] Thereby, for example, the description of "radius-ulna fracture" is expanded into "radius fracture" and "ulna fracture". Also, the description of "distal shaft fracture of radius" is expanded into "distal end fracture of radius" and "shaft fracture of radius". Also, the description of "C1, 2 fracture" is expanded into "C1 fracture" and "C2 fracture". Note that, as will be described later, during the repetition of the processes of steps S202 to S228, "C1 fracture" is converted to "atlas fracture" in the subsequent string conversion process (step S214). Similarly, "C2 fracture" is converted to "axis fracture" in the string conversion process (step S214).

[0048] And when there is a change in the expansion process (step S222), that is, when expansion is performed (Y in step S224), the unifier 102 advances the process to step S202 and performs the processes after step S202 with the string after expansion as the target string. In this way, the string conversion process is applied after the expansion process. Thereby, for example, it is possible to prevent "C1, 2 fracture" from being converted only to "atlas fracture" or only to "axis fracture".

[0049] The expansion process is performed after the partial matching process. For example, suppose the target string is "back first-degree burn". This string is registered as a standard medical term. On the other hand, its decomposition, "back burn" and "first-degree burn", are also registered as standard medical terms. Therefore, if the expansion process is performed before the partial matching process, "back first-degree burn" will be decomposed into "back burn" and "first-degree burn" against the author's intent. In contrast, in this embodiment, the expansion process is performed after the partial matching process, so it is possible to prevent such an expansion from occurring against the author's intent.

[0050] Furthermore, the expansion process is performed after the string conversion process. For example, suppose the target string is "Second-degree burns (TBSA 48%)". In this case, during the string conversion process, "TBSA 48%" in the target string is converted to "of 40-49% of body surface area", and the target string is converted to "Second-degree burns of 40-49% of body surface area". Then, during the expansion process, it is converted to "Second-degree burns, burns of 40-49% of body surface area". On the other hand, if the expansion process is performed before the string conversion process, it will be split into "Second-degree burns" and "TBSA 48%", resulting in an interpretation contrary to the author's intent.

[0051] If there are no changes in step S224 (N in step S224), the unification unit 102 proceeds to step S226. In step S226, the unification unit 102 performs extraction processing. In the extraction processing (step S226), the unification unit 102 searches for parentheses, etc., in the target string and extracts the text if the characters inside the parentheses match a disease name. For example, for the entry "Other (subarachnoid hemorrhage)", the characters inside the parentheses are removed, resulting in "Subarachnoid hemorrhage Other". In this case, the "subarachnoid hemorrhage" part is determined to be a partial match in the subsequent partial match processing (step S218).

[0052] Then, if there is a change in the extraction process (step S226), that is, if extraction has been performed (Y in step S228), the unification unit 102 proceeds to step S202 and performs the subsequent processing using the extracted string as the target string.

[0053] Furthermore, if there are no changes (N in step S228), the unification unit 102 proceeds to step S230. In step S230, the unification unit 102 performs a combination search. In the combination search process (step S230), the unification unit 102 attempts a partial match by sequentially delimiting the target string character by character.

[0054] For example, if the target string is "utsu kyūkyū dakinoku" (depression acute drug poisoning), the unification unit 102 splits it into "u" and "tsu kyūkyū dakinoku". The unification unit 102 treats each of the two resulting strings as target strings and determines whether each matches a standard medical term. If neither string matches a standard medical term, the unification unit 102 changes the delimiter position. This splits it into "utsu" and "kyūkyū dakinoku". Then, the unification unit 102 treats each string as a target string and determines whether each matches a standard medical term. In this way, the unification unit 102 repeats the process, changing the delimiter position one character at a time, until it matches a standard medical term. If, in the combination search, some of the strings match a standard medical term (Y in step S232), the unification unit 102 repeats the process from step S202 for the remaining strings. If there are no matches in the combination search, the unification unit 102 completes the medical term unification process.

[0055] As described above, in the medical terminology standardization process, the standardization unit 102 first determines whether the target string is an exact match to a standard term. Then, the standardization unit 102 applies multiple conversion rules, such as line splitting, numerical expansion, word conversion, partial matching, expansion, extraction, and combination search, in a predetermined order of application. In this way, the standardization unit 102 converts the target string appropriately according to each conversion rule and determines again each time whether it matches a standard medical term. This ensures that the target string is correctly converted to the corresponding standard medical term.

[0056] In the above, the transformation rules for combined searches are applied after the rules for partial matches. Therefore, it is possible to prevent unnecessary combined searches from being applied to medical terms that partially match.

[0057] Next, the standard format unification process (step S104), which was explained with reference to Figure 2, will be described. The standard format unification process is a process that unifies the data format of each data contained in the medical data into a predetermined standard format. The unification unit 102 unifies each data contained in the medical data into a predetermined standard format.

[0058] The reference format includes a static format that does not include time, and a time-series format that includes time. Patient information is static information that does not include time, and is recorded in the learning data DB 111 as learning data in static format. Treatment information and measurement results are information that corresponds to time, i.e., dynamic information that includes time, and are recorded in the learning data DB 111 as learning data in time-series format.

[0059] Furthermore, time series formats are classified into point time series formats and interval time series formats. Point time series formats are data formats for evaluation that are managed based on predetermined points in time. Information in point time series formats includes information on oral medication administered by patients, information on the administration of injectable drugs, etc. Here, information on oral medication refers to information that a certain drug was administered orally at a certain time. Information on the administration of injectable drugs refers to information that a certain drug was administered in injection form at a certain time. Other examples of information in point time series formats include vital signs (pulse rate, blood pressure, respiratory rate, etc.), blood tests (white blood cell count, creatinine, urea nitrogen, etc.), and blood gas analysis (pH, arterial oxygen partial pressure, lactate level, etc.).

[0060] Regarding the administration of injectable drugs, the time required for administration may or may not be controlled. For example, with anesthetics, the time may be controlled to the extent of slow administration, while with antibiotics, the administration time may be precisely controlled, such as administering them over one hour at 6:00, 14:00, and 22:00 each day. In this case, for anesthetics, a point time series format is preferable, and for antibiotics, a segment time series format is preferable. Thus, for the administration of injectable drugs, the time series format, whether a point time series format or a segment time series format, is predetermined depending on the type of drug administered and the method of administration.

[0061] Interval time series format is a data format for evaluation that is managed based on a predetermined interval. Information in interval time series format includes information on administration by intravenous infusion, information on procedures for which a period has been specified by healthcare professionals, etc. Information on administration by intravenous infusion indicates that administration by intravenous infusion was continuously performed over a certain interval (the period from the first time point to the second time point). Information on procedures for which a period has been specified indicates that a ventilator was used during a certain interval. Procedures include the use of medical devices such as ventilators. Examples of such medical devices include IABP (Intra-Aortic Balloon Pumping) and ECMO (Extracorporeal Membrane Oxygenation). Other examples of information in interval time-series format include intravenous fluid information (physiological saline, norepinephrine, aminoglycosides, etc.), ventilator settings (inspired oxygen concentration, pressure support, positive end-expiratory pressure, etc.), and non-invasive positive pressure ventilation (inspired oxygen concentration, pressure support, positive end-expiratory pressure, etc.). Thus, the evaluation target information that indicates each evaluation target related to the patient included in the medical data is standardized to a predetermined standard format for each type of evaluation target information.

[0062] In the standard format unification process (S104), the unification unit 102, for example, if numerical information for a ventilator is given in point time series format, rewrites it into interval time series format. On the other hand, the unification unit 102 manages information regarding oral administration and injection in point time series format.

[0063] For example, regarding drug administration, one medical institution's information system records the amount of drug administered every hour, while another medical institution's information processing system records the start and end times of drug administration and the rate of administration. However, in drug administration, it is preferable to record information that allows healthcare professionals to reproduce similar procedures, and such information is data for a specific period, such as what drugs are administered and over what period. Therefore, this information regarding drug administration is managed in a time-series format. On the other hand, information such as oral administration, injection administration, blood pressure measurements, and vital signs requires data at a specific point in time, and therefore, this information is managed in a point-time series format.

[0064] For example, the data sets shown in Figures 6 and 7 are treatment information related to intravenous drug administration, included in medical data transmitted from Hospital A and Hospital B, respectively. The treatment information from Hospital A shown in Figure 6 includes patient ID, time, order ID, bottle ID, status, drug, and dosage (ml), with the dosage recorded on a flow rate basis. Patient ID is the patient's identification information. Order ID is the order's identification information. Bottle ID is the drug bottle's identification information. Status is the status of the treatment.

[0065] On the other hand, the treatment information for Hospital B shown in Figure 7 includes patient ID, patient name, order ID, time, administration type, bottle ID, order type, drug ID, drug, solution volume, drug amount, and number of units. Only the administration amount is recorded, and data on flow rate is not included. Administration type is the type of administration. Order type is the type of treatment. Drug ID is the identification information of the drug. The two data sets also differ in the data they include, such as whether or not the patient's name and administration type are included. Furthermore, the treatment information for Hospitals A and B is associated with time, and the drug amount and other information at that time are recorded, making it point-time series information. Therefore, the unification unit 102 unifies all of this treatment information into interval time series format.

[0066] Figure 8 shows an example of the data structure of treatment information related to intravenous drug administration, recorded in interval time series format. In this embodiment, the unification unit 102 unifies the numerical data included in the medical data acquired by the acquisition unit 101, specifically the treatment information related to intravenous drug administration, into interval time series format.

[0067] In the interval time-series data structure, treatment information regarding intravenous drug administration includes administration ID, patient ID, start time, end time, administration rate, whether or not it is a bolus administration, drug name, and drug ratio. The administration ID is the identification information for the administration. The start time and end time are the start and end times of the procedure, respectively. Thus, in the interval time-series data structure, information such as the administration rate is managed for intravenous drug administration within the interval from the start time to the end time. In this way, in the interval time-series format, information regarding the evaluation target, intravenous drug administration, is recorded on an interval basis. The unification unit 102 unifies the treatment information regarding intravenous drug administration into the interval time-series data structure shown in Figure 8.

[0068] The data sets shown in Figures 9 and 10 represent treatment information regarding the use of ventilators, included in medical data transmitted from Hospital A and Hospital B, respectively. The treatment information from Hospital A shown in Figure 9 includes patient ID, time, instruction ID, status, item, and value. The instruction ID is the identification information for the treatment instruction. The status indicates the device being used. The item indicates the setting item, and the value indicates the value for that setting item.

[0069] The treatment information for Hospital B shown in Figure 10 includes patient ID, time, item name, and value. In all treatment information, information such as ventilation mode and oxygen concentration is managed for the interval from start time to end time.

[0070] Figure 11 shows an example of the data structure of treatment information related to the use of a ventilator, recorded in interval time-series format. In this embodiment, the unification unit 102 unifies the numerical data included in the medical data acquired by the acquisition unit 101, specifically the treatment information related to the use of a ventilator, into interval time-series format.

[0071] In the interval time-series data structure, treatment information related to ventilator use includes ventilator ID, patient ID, start time, end time, ventilation mode, oxygen concentration, positive end-expiratory pressure, inspiratory pressure, tidal volume, inspiratory time, respiratory rate, and inspiratory support pressure. Thus, the unification unit 102 unifies the treatment information related to ventilator use into the interval time-series data structure shown in Figure 11.

[0072] As described above, in the information processing system 1 according to this embodiment, the format of each data included in the medical data is unified into a predetermined standard format. In this way, it is possible to convert it into a format suitable for machine learning.

[0073] The embodiments described above are merely examples for carrying out the present invention, and various other embodiments can be adopted. For example, various modifications and changes are possible within the scope of the gist of the present invention as described in the claims, such as applying one modification to another. For example, some of the components of the above embodiments may be omitted, or the order of processing may be changed or omitted.

[0074] As a variation of this design, the processing performed by the model management device 10 in this embodiment may be implemented by multiple devices. That is, some functions of the model management device 10 may be implemented by a first device, and other functions of the model management device 10 may be implemented by a second device.

[0075] For example, the first device may perform processing up to the generation of a sudden change prediction model, and the second device may acquire the sudden change prediction model and perform sudden change prediction using the sudden change prediction model. Another example is that the model management device 10 may be provided integrally with a single hospital server 30. Furthermore, in this case, the medical data managed by the hospital server 30 may be used as training data to generate the sudden change prediction model.

[0076] According to the program, information processing system, and information processing method of this embodiment, which has the configuration described above, the acquisition unit 101 acquires multiple evaluation target information, each representing one of several types of evaluation targets related to a patient, and the unification unit 102 unifies the patient's evaluation target information into a standard format predetermined for each type. The standard format includes a static format that does not include time and a time-series format that includes time. The time-series format includes a point time-series format that shows the evaluation target at a predetermined point in time and an interval time-series format that shows the evaluation target in a predetermined interval. In this way, the information contained in the medical data is unified into a predetermined standard format, that is, a format suitable for machine learning.

[0077] Furthermore, according to the program, information processing system, and information processing method of this embodiment, the learning unit 103 may learn a predictive model to predict the patient's condition using evaluation target information in a standard format as learning data.

[0078] Furthermore, according to the program, information processing system, and information processing method of this embodiment, the point time series format may be set as the reference format for information regarding oral medication administered by a patient. Furthermore, according to the program, information processing system, and information processing method of this embodiment, the point time series format may be set as the reference format for information regarding the administration of injectable drugs to a patient. Furthermore, according to the program, information processing system, and information processing method of this embodiment, the interval time series format may be set as the reference format for information regarding the administration of injectable drugs to a patient. This makes it possible to obtain training data suitable for machine learning.

[0079] Furthermore, according to the program, information processing system, and information processing method of this embodiment, information regarding administration by intravenous injection to a patient may be set in interval time series format as the reference format. Also, according to the program, information processing system, and information processing method of this embodiment, information regarding a procedure for which a period has been specified by a medical professional may be set in interval time series format as the reference format. This makes it possible to obtain training data suitable for machine learning.

[0080] Furthermore, according to the program, information processing system, and information processing method of this embodiment, the information in interval time series format may include information that can reproduce the treatment specified by the medical professional. This makes it possible to obtain training data suitable for machine learning.

[0081] Furthermore, according to the program, information processing system, and information processing method of this embodiment, the measured values ​​in the treatment may be set in point time series format as the reference format. This makes it possible to obtain training data suitable for machine learning.

[0082] Furthermore, according to the program, information processing system, and information processing method of this embodiment, a point-time series format may be set as the reference format for information indicating a patient's vital signs. This makes it possible to obtain training data suitable for machine learning.

[0083] <Note> The matters described in the above embodiments are noted below.

[0084] (Note 1) In a computer (model management device 10) equipped with a processor (control unit 100) and a memory (storage unit 110), the processor (control unit 100) executes a program stored in the memory (storage unit 110) to acquire multiple evaluation target information indicating each of multiple types of evaluation targets relating to a patient, unifies the patient's evaluation target information into a standard format set in advance for each type, the standard format includes a static format that does not include time and a time-series format that includes time, the time-series format includes a point time-series format indicating an evaluation target at a predetermined point in time and an interval time-series format indicating an evaluation target in a predetermined interval.

[0085] Such computers can convert medical data, which is recorded in different formats by each medical institution, into a unified standard format such as static format, point time series format, or interval time series format. This eliminates problems caused by differences in data format and makes it possible to build a high-quality database that is easy to use for machine learning and other applications.

[0086] (Note 2) In the computer (model management device 10) described in Note 1, the processor (control unit 100) further learns a predictive model to predict the patient's condition by executing a program stored in the memory (storage unit 110), using the standard format evaluation target information as training data.

[0087] Such computers, by using evaluation data in a unified format, can efficiently train more accurate predictive models. This can improve the accuracy of predicting changes in a patient's condition, such as a drop in blood pressure or a decrease in oxygen saturation.

[0088] (Note 3) In the computer (model management device 10) described in Note 1 or 2, the processor (control unit 100) further learns a predictive model to predict sudden changes in a patient's condition by executing a program stored in the memory (storage unit 110) and using the standard format evaluation target information as training data.

[0089] Such computers can use standardized evaluation data as training data to train models that can accurately predict sudden changes in a patient's condition. This makes it possible to help prevent serious situations that would require admission to the intensive care unit.

[0090] (Note 4) In the program executed by the computer (model management device 10) described in any one of Notes 1 to 3, the information regarding oral medication taken by the patient is set in point-time series format as the standard format.

[0091] Such a computer can appropriately handle information about medication, which is an event at a specific point in time, as a point-time series. This allows for the generation of training data to accurately analyze the relationship between medication timing and changes in the patient's condition.

[0092] (Note 5) In a program executed by the computer (model management device 10) described in any one of Notes 1 to 4, the information regarding the administration of injectable drugs to patients is set in point-time series format as the standard format.

[0093] Such computers can concisely and appropriately represent the administration of injectable drugs, such as anesthetics, where the administration time is not strictly controlled or is completed in a short period of time, in a point-time series format. This allows for the acquisition of training data suitable for machine learning.

[0094] (Note 6) In a program executed by the computer (model management device 10) described in any one of Notes 1 to 5, the information regarding the administration of injectable drugs to patients is set in interval time series format as the standard format.

[0095] Such computers can accurately represent the administration of injectable drugs, such as antibiotics, which are administered continuously over a specific period of time, in a time-series format. This allows for more detailed analysis that takes into account the effects during the administration period.

[0096] (Note 7) In a program executed by the computer (model management device 10) described in any one of Notes 1 to 6, the information regarding administration to patients by intravenous injection is set in interval time series format as the standard format.

[0097] Such computers can treat procedures involving continuous administration, such as intravenous infusions, as interval information from start time to end time. This allows for data conversion without losing important information such as flow rate and administration duration, contributing to improved accuracy of machine learning models.

[0098] (Note 8) In a program executed by the computer (model management device 10) described in any one of Notes 1 to 7, information regarding a procedure for which a period has been specified by a medical professional is set in interval time series format as the standard format.

[0099] Such computers allow for the unified management of procedures performed over specific periods, such as the use of ventilators, in a time-series format. This makes it possible to accurately learn the correlation between the duration of a procedure and changes in the patient's vital signs.

[0100] (Note 9) In the computer described in Note 7, the information in interval time series format includes information that can reproduce the treatment specified by the medical professional.

[0101] Such computers can include not only the duration of a treatment but also essential information for reproducing the treatment itself, such as the administration rate and drug ratio. This improves the quality of the training data and enables the generation of more realistic and accurate predictive models.

[0102] (Note 10) In a program executed by the computer (model management device 10) described in any one of Notes 1 to 8, the measured values ​​in the treatment are set in point time series format as the reference format.

[0103] Such computers can appropriately handle measurements obtained as snapshots at specific points in time, such as blood pressure and blood test results, in a point-time series format. This allows for the accurate digitization and analysis of a patient's condition as it changes moment by moment.

[0104] (Note 11) In a program executed by the computer (model management device 10) described in any one of Notes 1 to 10, the point-time series format is set as the standard format for information indicating the patient's vital signs.

[0105] Such computers can uniformly handle continuously monitored vital signs such as pulse rate, blood pressure, and respiratory rate in a point-time series format, representing data at each measurement point. This allows for detailed tracking of changes in a patient's condition over time and enables learning to detect signs of sudden deterioration early.

[0106] (Note 12) In an information processing system (information processing system 1) comprising an acquisition unit (acquisition unit 101) and a unification unit (unification unit 102), the acquisition unit (acquisition unit 101) acquires multiple evaluation target information indicating each of multiple types of evaluation targets relating to a patient, and the unification unit (unification unit 102) unifies the patient's evaluation target information into a standard format predetermined for each type, the standard format includes a static format and a time-series format including time, and the time-series format includes a point time-series format indicating an evaluation target at a predetermined point in time and an interval time-series format indicating an evaluation target in a predetermined interval.

[0107] This type of information processing system allows medical data, which is recorded in different formats by each medical institution, to be converted into a unified standard format such as static format, point time series format, or interval time series format. This eliminates problems caused by differences in data format and makes it possible to build a high-quality database that is easy to use for machine learning and other applications.

[0108] (Note 13) An information processing method performed by a computer (model management device 10) having a processor (control unit 100), wherein the processor (control unit 100) acquires multiple evaluation target information indicating each of multiple types of evaluation targets relating to a patient, the processor (control unit 100) unifies the patient's evaluation target information into a pre-set standard format for each type, the standard format includes a static format and a time series format including time, and the time series format includes a point time series format indicating an evaluation target at a predetermined point in time and an interval time series format indicating an evaluation target in a predetermined interval.

[0109] This information processing method allows medical data recorded in different formats by each medical institution to be converted into a unified standard format, such as static format, point time series format, or interval time series format. This eliminates problems caused by differences in data format and makes it possible to build a high-quality database that is easy to use for machine learning and other applications.

Claims

1. A computer comprising a processor and memory, wherein the processor, by executing a program stored in the memory, acquires a plurality of evaluation target information representing a plurality of evaluation target items relating to a patient, unifies the evaluation target information of the patient into a predetermined standard format for each type, the standard format includes a static format that does not include time and a time series format that includes time, and the time series format includes a point time series format that shows an evaluation target at a predetermined point in time and an interval time series format that shows an evaluation target in a predetermined interval.

2. The computer according to claim 1, wherein the processor further learns a predictive model for predicting the patient's condition by executing a program stored in the memory, using the evaluation target information in the standard format as training data.

3. The computer according to claim 1 or 2, wherein the processor further learns a predictive model for predicting sudden changes in a patient's condition by executing a program stored in the memory and using the evaluation target information in the standard format as training data.

4. The computer according to any one of claims 1 to 3, wherein the point-time series format is set as the reference format for information regarding oral medication taken by a patient.

5. The computer according to any one of claims 1 to 4, wherein information relating to the administration of an injectable drug to a patient is set with the point-time series format as the reference format.

6. The computer according to any one of claims 1 to 5, wherein information regarding the administration of an injectable drug to a patient is set with the interval time series format as the reference format.

7. The computer according to any one of claims 1 to 6, wherein information relating to administration to a patient by intravenous injection is set with the interval time series format as the reference format.

8. The computer according to any one of claims 1 to 7, wherein information relating to a procedure for which a period has been specified by a medical professional is set as the interval time series format.

9. The computer according to claim 7, wherein the information in the interval time series format includes information that can reproduce a procedure specified by a medical professional.

10. The computer according to any one of claims 1 to 8, wherein the measured values ​​in the procedure are set in the point time series format as the reference format.

11. The computer according to any one of claims 1 to 10, wherein the point-time series format is set as the reference format for information indicating a patient's vital signs.

12. An information processing system comprising an acquisition unit and a unification unit, wherein the acquisition unit acquires a plurality of evaluation target information indicating each of a plurality of types of evaluation targets relating to a patient, the unification unit unifies the evaluation target information of the patient into a predetermined standard format for each type, the standard format includes a static format and a time series format including time, and the time series format includes a point time series format indicating an evaluation target at a predetermined point in time and an interval time series format indicating an evaluation target in a predetermined interval.

13. An information processing method performed by a computer having a processor, comprising: the processor acquiring a plurality of evaluation target information indicating each of a plurality of types of evaluation targets relating to a patient; the processor unifying the evaluation target information of the patient into a predetermined standard format for each type; the standard format comprising a static format and a time series format including time; and the time series format comprising a point time series format indicating an evaluation target at a predetermined point in time and an interval time series format indicating an evaluation target in a predetermined interval.

Images