Method for generating a query message with a prompt and method for generating a response message to a query message

Abstract

The invention relates to a computer-implemented method for generating a query message (QM) with a prompt in an information request system, wherein the prompt is configured to be input into a source language model of an information source system that is separate from the information request system, in order to cause the source language model to provide missing medical information (MI). The method comprises the following steps: - Obtaining (S10) a patient cache (22) corresponding to a patient of interest, wherein the patient cache (22) contains medical information (MI) about the patient of interest; - Delivering (S20) the received patient cache (22) as context to a demand language model; - Generate (S30), using the request language model and based on the patient cache (22), a query message (QM) with a prompt, wherein the prompt is configured to be input into a source language model of at least one information source system to cause the source language model to provide missing medical information (MI) about the patient of interest; - Obtain (S40) an information source contact address of at least one information source system; - Sending (S50) the generated query message (QM) to the at least one information source system using the information source contact address.

Description

[0001] The invention relates to a computer-implemented method for generating a query message with a prompt. Furthermore, the invention relates to a computer-implemented method for generating a response message to a query message. The invention also relates to a corresponding data processing device, a computer program product, and a computer-readable storage medium.

[0002] Regardless of the grammatical usage of the term, persons with male, female or other gender identities are included within the term.

[0003] To provide optimal patient care, it is essential to have important medical information readily available when needed. This enables healthcare providers to make informed decisions, such as regarding a patient's further treatment. However, medical information is often scattered across various databases, documents, or institutions. Therefore, accessing medical information is difficult, and searching for missing information is cumbersome. For example, a radiologist who needs background information to make a diagnosis may have to contact the referring physician or the laboratory department to obtain relevant information about a patient's medical history, symptoms, previous diagnoses, or test results.This process of gathering missing medical information can be time-consuming, cumbersome, and prone to errors or delays. This negatively impacts the quality and efficiency of healthcare.

[0004] Therefore, there is a need for a system that can automatically identify, locate, request, and deliver missing medical information. Such a system would allow healthcare providers to quickly and conveniently access the information they need without interrupting their workflow or requiring manual intervention.

[0005] Some progress has been made in making medical information automatically available to healthcare providers. For example, there are efforts to standardize data protocols, data structures, and data encoding (e.g., DICOM, HL7, FHIR, or LOINC). As a result, the situation has improved, and some systems are now able to communicate medical information automatically.

[0006] Nevertheless, many phone calls are still required to request medical information, such as documents, lab results, a current patient status, or a medication chart, from other departments or institutions. Other commonly used manual methods for requesting information include sending emails or faxes to request information from a potential source. These manual steps are time-consuming, distract healthcare providers from their clinical routine, and create a significant financial burden on the healthcare system.

[0007] The invention aims to overcome these limitations and to provide methods and systems that can support healthcare providers in retrieving missing medical information. The methods and systems provided must be easy to use and readily accessible. Furthermore, a healthcare system using the provided systems and methods should be easily scalable.

[0008] This task is accomplished by the features of the independent claims. Further advantageous examples are included in the dependent claims.

[0009] A computer-implemented procedure for generating a query message in an information request system is provided. The query message includes at least one prompt, which is configured to be input into a source language model of an information source system to induce the source language model to provide missing medical information. The information source system is separate from the information request system. The procedure comprises the following steps: - Obtaining a patient cache that matches a patient of interest, wherein the patient cache contains medical information about the patient of interest; - Providing the received patient cache as context for a requirements language model; - Generate, using the request language model and based on the patient cache, a query message with a prompt, wherein the prompt is configured to be input into a source language model of at least one information source system to cause the source language model to provide missing medical information about the patient of interest; - Obtaining an information source contact address of at least one information source system; - Sending the generated query message to at least one information source system using the information source contact address.

[0010] The information source system and / or the information request system can be a local pool of databases, with the databases of the information source system being separate from those of the information request system. This means that from outside either system, it can be difficult to identify what information is available in that system's databases. It can be even more difficult to retrieve missing information that is available in that system's databases from outside either system. In other words, the information request system and / or the information source system can be an information silo. For example, the information request system and the information source system could serve different healthcare organizations, different hospitals within a healthcare organization, different departments within a hospital, and so on.or even a combination of the aforementioned examples.

[0011] In other words, from a data exchange perspective, the information request system can be separate or isolated from the information source system. Regardless of the specific local arrangement of the information request system and the information source system, this means that both the information request system and the information source system each constitute a self-contained information silo. Each information silo contains medical data, with the medical data in one information silo not being easily, completely, or at all accessible from the other information silo, or vice versa. For example, the medical data / information in the information request system may be stored in a different location or in a different format than the medical data / information in the information source system.

[0012] The information request system and the information source system can be connected, at least temporarily, for data / information transfer and / or exchange. The systems can be connected via a shared information exchange connection, particularly for sending query messages and / or response messages. This information exchange connection can be implemented using a network connection. For example, the network could be a local area network (LAN), the internet, a private network such as an intranet, or a wide area network (WAN). The network connection is preferably wireless, e.g., a wireless LAN (WLAN or WiFi).

[0013] The information request system and / or the information source system may include various components of a health information system, such as image storage and processing components, e.g., a Picture Archiving and Communication System (PACS), and / or radiology reporting and / or review components, e.g., a Radiology Information System (RIS). The systems may include components or processes that can be set up for computerized provider order entry (CPOE), clinical decision support, patient monitoring, population health management (e.g., a Population Health Management System (PHMS), health information exchange (HIE), etc.), health data analytics, and / or cloud-based image sharing. The systems may also include electronic health record systems (e.g.,This includes systems for an electronic medical record (EMR), an electronic health record (EHR), an electronic patient record (EPR), a personal health record (PHR), etc., and / or other health information systems (e.g., clinical information system (CIS), hospital information system (HIS), patient data management system (PDMS), laboratory information system (LIS), cardiovascular information system (CVIS), etc.).

[0014] In one example, the information request system and / or the information source system includes at least one component for storing medical information, such as a HIS, RIS and / or PACS.

[0015] Medical information can include, for example, clinical reports, patient information, and / or administrative information received by staff in a hospital, clinic, and / or physician's office (e.g., an EMR, EHR, PHR, etc.). Other examples of medical information include radiology reports, radiology examination image data, messages, alerts, alarms, patient scheduling information, patient demographic data, patient tracking information, and / or physician and patient status monitors. Medical images (e.g., X-rays, scans, three-dimensional renderings, etc.) are considered another type of medical information.

[0016] Medical information can be stored in a database or registry, for example. The database can contain various subcomponents, i.e., multiple sub-databases. In some examples, medical images can be stored in a PACS, such as using the DICOM format (Digital Imaging and Communications in Medicine). Images can be stored in the PACS by medical professionals (e.g., imaging technicians, physicians, radiologists) after a patient has undergone medical imaging, and / or automatically transferred from medical imaging devices to the PACS for storage. Radiological data can be stored in a RIS. Patient data and clinical reports can be stored in a HIS.Therefore, in an exemplary design of the information source system and / or the information request system, the database of the system for storing the medical information may include a PACS, HIS and / or RIS, depending on the medical information available to be stored.

[0017] In general, "receiving" can include "collecting" and / or "determining" and / or "receiving" and / or "retrieving" and / or "querying a database to obtain" regardless of how the data / information / elements to be obtained are specifically implemented.

[0018] A patient cache can contain medical information about a patient. The patient cache can be a more accessible representation of the data / information available about a patient. Information about the patient can be easily retrieved from this patient cache by prompting the LLM (Large Language Model). The patient cache can be created by providing all relevant medical information about a patient as context for the language model. For example, the patient cache can be created by inputting medical information about a patient of interest into the language model.

[0019] In one exemplary implementation, the patient cache can be a chat history of the language model, including the prompts fed to the language model and the responses generated by the language model during a session. In other words, the patient cache can be a representation of the language model's knowledge about a patient. The patient cache can be fed to the language model by loading it into the language model's context, for example, using a "load chat history" function.

[0020] The language model can be used to filter out irrelevant medical information from a patient's available medical information or to summarize information, e.g., when the language model's context is too small to hold all of a patient's available medical information / data of interest.

[0021] Medical information about a patient is preferably converted into a format that can be processed by the language model. For example, if the medical information about a patient includes scanned documents, optical character recognition (OCR) can be applied to convert the information into machine-readable content.

[0022] The information request system and / or the information source system can include a standalone database, the requester-source patient cache database or the source patient cache database, in which the patient caches are stored. The patient caches can be saved by calling a "save chat history" function; preferably, each patient cache corresponds to a specific patient and contains medical information about that specific patient. As a result, the patient cache is easily accessible and can be loaded from the patient cache database without having to create a new patient cache. When a query message, i.e., a request for medical information about a patient, is received, for example,If the patient cache is received from an information request system, a doctor, or the like, it can be quickly provided as context for the language model.

[0023] The patient caches and patient cache database are comparable to an EHR, with the EHR being automatically kept up-to-date and missing information being automatically requested from other systems. The information request system itself can be queried by other systems or users to provide missing medical information, for example, from a HIS or RIS. It may include a chat interface and / or a chatbot that a healthcare provider can use to retrieve missing medical information in a conversational manner. This improves access to medical information for end users.

[0024] In one example, the patient cache can contain medical information in embedded form. "Embedding" can involve encoding the patient cache / medical information. In other words, the medical information can be translated into an array of numbers—that is, a vector, a matrix, a tensor, or the like—to facilitate processing of the medical information by the language model. Although the medical information is embedded, the embedding still expresses the original meaning of the data, so a certain similarity between embeddings indicates a similarity in the underlying data. Thus, expressing data in embedded form, especially as vectors, enables the interoperability of data originating from different sources, such as text data and / or image data.In particular, embedding the data in the same embedding space allows multimodal data to be processed by a single language model.

[0025] The embedded information can serve as a pointer to the location where the underlying medical information is stored. The storage capacity required to hold the embedded information can be less than the storage capacity required to hold the (raw / non-embedded) medical information. Embedding the information thus allows for efficient use of available storage capacity. For example, the embedded information can be stored in a location directly accessible to a computing unit performing the required procedure, requiring only minimal storage capacity. The non-embedded data can be stored in an external database with higher storage capacity, such as in a cloud.

[0026] The embedded medical information can be fed into the language model by enriching a prompt that is then fed to the language model containing the embedded medical information. For example, a retrieval-augmented generation (RAG) framework can be provided that is capable of providing context to the language model, with the context based on the medical information available about a patient of interest. After generating a query message or a response message with the language model, the sources on which the generated message is based can be presented to system users or included in the generated query or response message. This ensures that the model's responses can be checked for accuracy and correctness.Therefore, using a RAG framework reduces the risk of hallucinations.

[0027] In a preferred embodiment, the RAG framework includes two phases: a retrieval phase and a content generation phase.

[0028] During the retrieval phase, a patient cache containing embedded medical information is fed into the language model. For example, a patient cache database might contain a multitude of patient caches, each with embedded medical information about a specific patient. To identify the most relevant patient cache within the database, a vector search can be performed. The patient cache database can be searched for patient caches with embedded medical information, which most closely resembles a vector representation of the question / prompt / patient information input into the language model. Various search algorithms can be used for this, such as k-nearest neighbors (KNN) or hierarchical navigable small world (HNSW).

[0029] During the content generation phase, the language model is prompted to generate a query message or a response message. The prompt is augmented with the identified patient cache, i.e., the medical information contained within the patient cache. This allows the language model to generate an accurate response. The credibility of the response can be enhanced by providing links to its sources.

[0030] The language model, i.e., the request language model and / or the source language model, can comprise a deep neural network, for example, one that employs a transformer architecture. The language model can be a large language model (LLM) with a large number of neurons, such as one billion neurons or more, preferably seven billion neurons or more, most preferably thirteen billion neurons or more, and most preferably thirteen billion neurons or more. Specifically, the language model can be designed to receive a sequence of tokens, such as words, syllables, or lexemes, as input and output a token considered the "most likely" next token.In particular, the language model can be operated in a generative or autoregressive manner, that is: an entry, which is a sequence of tokens, can be fed to the language model as input; the first most probable next token output by the language model can be appended to the entry to obtain a new entry to be fed to the language model as input in the next iteration; and by repeating this sequence of iterations, the language model may generate a meaningful most probable response message to the entry, comprising several sentences or even paragraphs.

[0031] The invention particularly utilizes the universal information processing and communication capabilities of modern language model-based chatbots such as Chat-GPT or more specialized ones, such as Med-PALM 2. These language models allow for a large size of the delivered context, e.g., 32k tokens for GPT-4 from OpenAI, which is comparable to about 50 A4 pages in text size 11, or 100k tokens (about 150 pages) for Claude from Anthropic.

[0032] This allows the language model to be trained to provide medical information about a patient when prompted accordingly. This means that when a suitable prompt is entered into the language model, the model will provide medical information about a patient, retrieving this information from a patient cache loaded into the language model's context. For example, the language model can be trained to generate a response message containing medical information about a patient, specifically any missing medical information.

[0033] Additionally or alternatively, the language model can be trained to generate a query message including a prompt. The prompt can specify missing medical information about a patient and / or include a command that, when the prompt is input into a source language model, causes the source language model to provide the specified missing medical information and / or generate a response message including the specified missing medical information.

[0034] Furthermore, the language model can be trained to identify missing medical information in a patient cache and / or, based on a patient cache, to identify at least one potential information source system that can provide the missing medical information about a patient, as will be explained in more detail later.

[0035] In particular, the language model can be pre-trained to understand natural language and to provide meaningful answers on any given topic. Such pre-training is possible by using a text database, the result of crawling a portion, preferably a substantial portion, of the internet, as language training data. Therefore, no labeling of the language training data used for pre-training is required. However, pre-training the language model can optionally include supervised training steps using labeled language training data generated by humans to further improve the language model's ability to provide meaningful answers to the user.

[0036] Examples of pre-trained generative language models include OpenAI's ChatGPT 3 and 4, Google AI's Palm, BERT, and Gemini, DeepMind's Chinchilla, and Meta's Llama 1, 2, and 3 models. Meta's latter pre-trained language models are available for download from the internet, whereas the earlier pre-trained models can be accessed and licensed online, including the option for further customized fine-tuning.

[0037] The language model can be trained from scratch or by fine-tuning an existing model, for example, using supervised, unsupervised, or semi-supervised learning techniques. In particular, the language model can be fine-tuned according to the proposed solution for the tasks to which it is to be applied. That is, fine-tuning can involve further unsupervised or supervised training of the language model using language training data related to the proposed system.

[0038] For example, the language model can be trained using a labeled training dataset. The labeled training dataset can contain medical information about a patient, with missing pieces of medical information being identified and the training dataset examples labeled accordingly.

[0039] The language model can be trained using backpropagation techniques. For example, different entries (e.g., prompts, training samples from patient caches) can be fed into the language model in a training dataset, and output can be generated by the language model; for example, the language model can extract medical information from a patient cache.

[0040] A difference between the output of the language model and a label (or identifier) ​​for the entry can be determined according to a loss function. Based on this difference, backpropagation techniques can be used to adjust the parameters and / or weights of the language model. The language model can thus be trained over time with various labeled entries, for example, until the language model is accurate to a certain threshold. At this point, the language model can be used in one of the claimed methods.

[0041] For example, the training dataset can be segmented into three groups (i.e., training, validation, and testing). The language model can be trained based on the first group (training). The model can then be run (at least partially) to predict results for the second group of data (validation). Using the procedure described above, it can be evaluated whether the language model has been trained correctly. The accuracy of the model can be measured using the remaining data points within the training dataset (testing) (e.g., area under the curve, precision, and recall).

[0042] The language model preferably takes a prompt as input and generates its output based on the information provided to the language model as context.

[0043] The language model can be stored, at least partially, within the premises of the information request system and / or the information source system, such as within the IT infrastructure of a hospital. In particular, each system can utilize its own independent language model. Thus, the information request system can utilize a request language model, and / or the information source system can use a source language model. Since the data fed into the language model does not leave the organization's premises in this case, privacy is guaranteed. Additionally, or as an alternative, the language model can be stored in a cloud and accessible via the internet; for example, the language model can be shared. This means that the request language model and the source language model are, in fact, the same model.The language model can be prompted from within the information request system and / or the information source system in such a way that the data sent to the language model leaves the premises of the information request system and / or the information source system. In this case, it may be advantageous to pseudonymize or anonymize medical information / data fed to the language model.

[0044] The prompt that may be fed to the language model can describe the task or command that the language model is to perform or execute. The prompt is preferably based on natural language. In this disclosure, a prompt is an instruction signal or input that is fed to the language model to initiate a specific response or action.

[0045] For example, entering a prompt into a source language model can cause the source language model to provide missing medical information that may be needed to fill an information gap in the patient cache at the information request system. The appropriate prompt could be, for example, "Provide... about patient XY to allow an informed decision about the next steps in the patient's clinical pathway!", "What imaging data is available for patient XY?", "Send me the latest X-ray as an ASCII-encoded DICOM!", "When was the last complete blood count performed and what are the lab results?", or other commands or questions.

[0046] Additionally or alternatively, a prompt can cause the request language model to identify missing information in a patient cache when the prompt is input into the request language model. The corresponding prompt might be, for example: "What additional information might be helpful in making an informed decision about the next steps in the patient's clinical pathway?"

[0047] Additionally or alternatively, a prompt can cause the request language model to identify at least one potential source for the missing medical information, i.e., at least one potential information source system that can provide the missing medical information. A corresponding prompt might then be, for example: "Find the contact address of an information source from which the required additional information can be obtained."

[0048] The prompt that causes the source language model to provide missing information about the patient of interest can be automatically generated by the request language model. For this purpose, an automatic prompter interface can input a corresponding prompt into the request language model, which in turn causes the request language model to generate a prompt configured to be input into the source language model. The automatically generated prompt can provide background information to the potential information source, which the source language model can then use to more precisely identify the missing information.Background information can also be used to identify additional information that may not be explicitly requested by the information request system in the prompt, but which may nevertheless be useful in combination with the missing medical information, e.g., to determine the next steps in a patient's therapy.

[0049] By automatically generating a query message with a prompt configured to be entered into a language model to prompt the language model to provide missing medical information, and automatically sending the query message to a potential information source system, manual queries by healthcare providers for missing information are replaced by automated queries. This frees healthcare providers from the time-consuming task of manually gathering information and allows them to focus on other issues, such as improving patient care or increasing diagnostic accuracy.

[0050] In a preferred embodiment, the method further comprises the step: - Identify, using the request language model and based on the patient cache, missing medical information in the patient cache; and / or - Identify, using the requirements language model and based on the patient cache, at least one potential information source system that can provide the missing medical information.

[0051] In principle, information about missing medical information and / or the identification of a potential source for that missing medical information can be provided by a human operator. However, automatically identifying missing medical information based on the patient cache and automatically identifying at least one potential information source system from which the missing medical information can be provided saves additional time for healthcare professionals. Furthermore, once the language model is properly trained to perform the described procedural steps, the process is largely independent of the user's skill level. Thus, the system can be operated by professionals with a wide range of skill levels while maintaining consistent quality of results.

[0052] As explained above, the request language model and / or the source language model can be operated via an automatic prompter interface in the information request system and / or the information source system. For example, the automatic prompter interface can determine which prompts should be entered into the language model, in what order the prompts are entered, what happens to the generated results, etc. In one example, the automatic prompter interface is rule-based. This means that the actions of the automatic prompter interface are determined based on a multitude of if-then rules. Operating the automatic prompter interface thus requires few resources. Alternatively, the automatic prompter interface can be implemented as a language model.

[0053] In one embodiment, the automatic prompter interface can input a prompt into the request language model to identify missing information in the patient cache of the patient of interest. The prompt can be specific to a future action the patient will undergo, such as surgery. A specific piece of information to be retrieved can be identified, such as an image of a particular region of the patient's body. Alternatively, the prompt can be more general, for example, asking what information might be useful for planning the patient's further treatment. To identify the missing medical information, the language model can extract a use case for the information from the prompt. The language model can analyze the available information contained in the patient cache.The language model can compare the available information with the information required in examples with a similar use case. This allows it to identify which information is missing from the patient cache.

[0054] In one example, the missing information might be needed by another artificial intelligence system, such as an image processing system, to enable it to accurately analyze the clinical condition of a patient of interest. Based on this additional information, the insights generated by the other AI system can be improved. For instance, the missing medical information could be used by an image processing system designed to analyze a medical image and generate insights based on that image, thereby improving the quality, consistency, and / or reliability of the insights it produces.

[0055] The missing medical information can be helpful in better understanding and / or analyzing a patient's clinical condition and / or verifying findings. For example, the missing medical information can be used to check whether a patient's symptoms are consistent with the findings. In another example, a CT scan with contrast medium might be scheduled, requiring prior examination of the patient's organs, such as the liver or kidneys. In this case, the missing medical information may be related to, or constitute, previous tests of the functionality of these organs.

[0056] Preferably, once the missing information has been identified, the automatic prompter interface can inject a prompt into the request language model to identify at least one potential information source that might provide the missing medical information. For example, the request language model can analyze the medical information in the patient cache, searching for the address of a primary care provider, such as a general practitioner, a laboratory, or any other health information silo. Potential information sources might also be found on a laboratory request, a radiology request / order form, a pathology request, or a general practitioner's discharge notes.

[0057] In some cases, important information source systems can also be explicitly added to the patient cache to guarantee that these information source systems are detectable by the requirements language model.

[0058] Preferably, the step of obtaining an information source contact address involves identifying the information source contact address of the potential information source system using the request language model, based on the received patient cache. The request language model automatically extracts an information source contact address from the at least one information source system. This could be an email address, another information exchange protocol, or a dedicated information exchange API. Additionally, or as an alternative, a lookup table can be provided that stores the contact addresses for the information sources, and the information source contact address can be retrieved from the lookup table once the potential information source has been identified.

[0059] In a preferred embodiment, the method is executed at regular intervals, e.g., once a week, and / or on demand, e.g., when a user of the system requires additional information about a patient. Additionally, or alternatively, the method can be executed when new medical information about a patient of interest becomes available. This additional, new medical information can then be added to the patient cache of the corresponding patient of interest. The additional medical information may allow further inferences to be drawn about the patient. However, some medical information may be missing before these inferences can be made.By automatically executing the claimed method, missing information can be automatically retrieved as soon as the additional information becomes available, enabling professionals to obtain a complete picture of the patient, even if they are encountering the additional information for the first time.

[0060] Once a query message has been generated, it can be sent to the identified information source system. The following aspects, concerning the generation of a response message to a query message, can be performed independently of the query message generation or in combination with it, for example, directly after the query message has been generated.

[0061] Another aspect of the invention relates to a computer-implemented method for generating, in an information source system, a response message to a query message from an information request system separate from the information source system, wherein the method comprises the following steps: - Received, in the information source system, a query message from an information request system, wherein the query message includes at least one prompt, the prompt being configured to be input into a source language model of the information source system to cause the source language model to provide missing medical information about a patient of interest; - Providing a source patient cache database containing a multitude of patient caches, with each patient cache containing medical information about a patient; - Obtained, from the source patient cache database, at least one patient cache that matches the patient of interest; - Feeding the received patient cache as context to the source language model of the information source system; - Generating a response message to the query message by entering the prompt into the source language model, where the response message at least specifies: - if the missing medical information can be provided based on at least one patient cache obtained, the missing medical information, or, - if the missing medical information is not available based on the at least one patient cache received, that the missing medical information is not available in the information source system; - Obtaining a requester contact address for the information request system; - Sending the generated response message to the information request system using the requester's contact address.

[0062] As mentioned, at least parts of the method for generating a response can be executed directly after the method for generating a query message. This means that parts or all of the claimed methods can be combined.

[0063] The information source system can identify the patient of interest based on the information provided in the query message. The query message may include a patient identifier, such as a unique patient ID and / or a patient name, to enable identification of the patient of interest within the information source system.

[0064] If a patient cache corresponding to the patient of interest is available in the patient cache database, the patient cache can be loaded from the database and entered as context into the source language model. If no patient cache is available, a new patient cache can be created for the patient based on the information contained in the query message. Checking whether a suitable patient cache is available can be performed, for example, through the automatic prompter interface of the information source system or through the source language model itself.

[0065] At least the prompt contained in the query message can be input into the source language model, for example, via the information source system's automatic prompter interface, to generate a response from the source language model. Alternatively, the entire message, including the prompt, can be input into the source language model for processing. The source language model can identify the prompt in the query message and generate a response based on the prompt and the provided patient cache.

[0066] The response generated by the source language model can indicate, based on the information available in the information source system, whether the missing medical information is available. Specifically, the response can indicate whether the missing information is available in the information source system, particularly in the patient cache provided as context to the language model. Furthermore, the query message can include the missing medical information requested by the information request system, if this information is available in the information source system. If the requested medical information is not available based on the information available in the information source system, the generated response can indicate this accordingly.

[0067] There may be cases where the identity of the patient of interest is not fully apparent from the query message. For example, there may be multiple patients with the same name. Consequently, there may be multiple patient caches corresponding to the patient with the same name in the patient cache database. The additional information provided in the query message may not be sufficient to definitively determine which of the patients is the patient of interest. In this case, the information source system can search all patient caches that could be relevant to answering the query message. For example, the information source system can iteratively retrieve / load a patient cache from the patient cache database that corresponds to a potential patient of interest.The patient cache can be searched for relevant information, and a response can be generated regarding potentially missing information. Then, the next potentially relevant patient cache can be retrieved, and the loop continues until all potentially relevant patient caches have been searched and corresponding responses generated. At least one of the generated responses, preferably some, especially the most relevant one, and even more preferably all of the generated responses, can be included in the response message. Alternatively, multiple response messages can be generated, each based on at least one of the generated responses.

[0068] Each response message can be associated with information that allows for the reliable identification of the patients it contains. This means that the information source system can add information to the response message that was not included in the query message, in order to avoid confusion about the identity of the patients being processed in the response message. For example, if only the patient's name is included in the query message and more than one patient with that name is found in the information source system, the information source system can add the date of birth of each patient to the generated response to avoid misunderstandings.

[0069] In other words, an incoming query message can be processed by the information source's automatic prompter interface. The patient in question can be identified, and that patient's cache can be loaded. Finally, the prompt requesting the missing medical information, preferably along with additional information about the patient, can be forwarded to the speech model. The speech model is prompted to generate the requested response, and this response can be collected by the information source system's automatic prompter interface and sent back to the origin of the query message.

[0070] There may be instances where the language model detects that the missing information must be available somewhere in the information request system or the information source system. However, the information is not contained in the patient cache of the patient of interest. In other words, the language model may decide that the missing information is likely available but has not been fed into the patient cache corresponding to the patient of interest. In this case, the language model may send a manual information request to the staff of the information silo, i.e., the information request system or the information source system. For example, such a manual information request might appear on an interaction interface of the information request system or the information source system, in a specific mobile app, or a desktop application.Manual information requests can be prioritized based on their time criticality.

[0071] In a preferred embodiment, particularly in the case that the missing medical information is not available in the information source system, the method may further include the following step: - Identify, using the source language model and based on the received patient cache and / or query message, at least one other potential information source system that may be able to provide the missing medical information; - Generating, using the source language model, a query message with a prompt, wherein the prompt is configured to be input into a source language model of at least one other information source system to cause the source language model to provide missing medical information about the patient of interest; - Obtaining an information source contact address of at least one other information source system; - Sending the generated additional query message to at least one other information source system using the information source contact address.

[0072] In one embodiment, the identification of further potential information sources and the sending of further query messages to these additional potential information sources can occur if, for example, the missing medical information requested in the query message is not available in the information source system and / or if, for example, the information source system determines, based on the received patient cache and the query message, that another potential information source system may be better suited to providing the information or may provide additional information and / or information of a higher quality.

[0073] The source language model can be used to identify potential additional sources for the missing patient information. This can be done based on the received query message and / or the patient cache retrieved from the source patient cache database. Since the patient cache in the information source system may contain different medical information than the patient cache in the information request system, the information delta can be used to find other or more promising sources for the missing medical information. Thus, generating and sending the additional query message can be considered a second iteration of the query process. In this second iteration, the information source system from the first iteration can be considered the information request system in the second iteration.

[0074] The source language model is thus used to generate another query message with a prompt. In other words, the source language model serves as the request language model. The generated query message is sent to another potential information source system and processed there, so that the language model of that other potential information source system serves as the source language model.

[0075] The response message generated in a further information source system, i.e., a response message generated in the second or a subsequent iteration of the query process, can be sent directly to the original information request system, i.e., the information request system of the first iteration. Additionally, or alternatively, the response can be sent to the information source system from which the further query message was received; i.e., in the case of a second iteration query, the information source system from which the query message was sent.

[0076] By performing several iterations of the claimed method, a network of multiple information systems can be searched efficiently, in particular in a snowball-like manner.

[0077] Additionally, or as an alternative, further query messages can be generated and sent by the information request system. In a preferred embodiment, the method thus further comprises the following steps: - Receiving a response message from the information source system indicating that the missing medical information is not available in the information source system; - Identify, using the request language model and based on the patient cache, at least one other potential information source system that can provide the missing medical information; - Generate, using the request language model and based on the patient cache, at least one further query message with a prompt, the prompt being configured to be input into a source language model of the further information source system to cause the source language model to provide the missing medical information about the patient of interest; - Obtaining an information source contact address from the broader information source system; - Sending the generated query message to the further information source system using the information source contact address.

[0078] For example, this process can be used if, in the first iteration, for reasons of procedural efficiency, only a query message was sent to a single information source system, but several information source systems have already been identified from which the missing medical information can be obtained. Then, the identified potential information source systems can be contacted sequentially by the information request system, for example, by iterating through a list of identified potential information source systems, until one of the queried information source systems sends an appropriate response message containing the missing medical information.

[0079] As explained above, the query message can be generated by the request language model and / or by the source language model. Likewise, the response message can be generated by the source language model. The content of the query message and / or the response message can vary, but the query message must include at least one prompt that could be used to request the missing medical information from the language model to the target information source system, as specified above.

[0080] For example, the query message and / or the response message can contain information about the patient of interest, such as a unique identifier like name, date of birth, patient ID, etc., which allows the information source system to preferentially and unambiguously identify the patient of interest. The query message and / or the response message can also include authentication information, a unique query ID, and / or the timeframe within which the missing information is needed / was provided.

[0081] The following explanations focus on the characteristics of query messages to keep them concise and easy to read. However, these explanations and the described characteristics also apply to the generated responses / response messages.

[0082] In a preferred embodiment, the query message includes a unique query ID, and the method further comprises the following steps: - Obtain a current query ID from the query message to which a response should be made; - Checking whether the query message has already been answered by comparing the current query ID with query IDs in a query ID list, where the query ID list contains query IDs of query messages that were previously answered in the information source system; - if the query message has already been answered, the procedure will be terminated.

[0083] As a result, a note can be sent to the sender of the query message, e.g., to the information request system, indicating that the query message has already been answered.

[0084] The unique query ID can be a unique hash, allowing identification of the system in which it was generated. The hash can be a multi-stage hash, such as a two-part hash. Hash generation can be based on a hash function. The system can include a central query ID provider, such as a central server, responsible for mapping unique query IDs to query messages and ensuring their uniqueness. Additionally, or alternatively, each system—that is, each information request system or information source system—can include a unique query ID generator capable of generating unique query IDs. For example, the unique query ID generator can produce a unique hash that can be included in a query message.

[0085] Similarly, a unique query ID can be included in the responses generated in the information source systems, particularly in the response messages.

[0086] The query message and / or the response message may preferably contain authentication information, the authentication information being designed to be verified in the information request system or in the information source system. Preferably, the authentication information is verified immediately upon receipt of a query message and / or a response to a query message, i.e., before the further process of delivering missing medical information or processing missing medical information continues.

[0087] The authentication information can be a token issued by an authentication system, particularly a centralized or decentralized authentication system. This token can be verified by the information system that receives the query message or the response message.

[0088] Additionally, or as an alternative, the authentication information can be based on a system of hashes derived from previously exchanged information. The authentication information preferably includes at least one hash of previously exchanged information. The hash can correspond to a query message or a response message previously sent between the information requesting system and the information source system. The hashes of exchanged query messages and / or responses can be stored in the information requesting system and / or the information source system. Thus, the receiving information source system and / or the receiving information requesting system can compare the hash sent with the current communication with previously sent hashes to verify the provided authentication information.In one example, the authentication information could be the unique query ID. The hash that serves as the unique query ID can also be used as authentication information.

[0089] The authentication process is preferably carried out via the Automatic Prompter interface.

[0090] In a preferred embodiment, the requesting contact address and / or the information source contact address is an email address. Email addresses are widely available and therefore facilitate the connection of different healthcare information systems. A specific tag can be included in the email subject line (e.g., "FAXIQREQ"). Using such a tag, emails exchanged between an information request system and an information source system to obtain missing patient information can be automatically redirected. For example, the emails can be sent directly to the automatic prompter interface of a receiving system without interfering with other email communication between the information request system and / or the information source system.

[0091] In this case, the request message can be in the form of an email. Additionally, or alternatively, the response message can also be in the form of an email. The email can include a subject and body. Emails can be sent via a communication interface, which may be an email server of the information request system or the information source system. Depending on its origin, the email can be generated by the request language model or the source language model.

[0092] Inquiry messages and / or reply messages can be sent via an encrypted communication channel, e.g., via TLS-encrypted email, to guarantee patient privacy and to protect patient information during transmission.

[0093] As briefly explained above, the query message, the response message, and / or the patient cache can contain a patient identifier that corresponds to the patient of interest. The patient identifier can allow for the unambiguous identification of the patient of interest. The patient identifier can be a unique identifier, a name, a date of birth, a patient ID, a place of origin, etc., or a combination thereof.

[0094] To build the patient cache database, the procedure may preferably also include the following steps: - Obtaining medical information about a patient of interest; - Check if a patient cache is available for the patient of interest in a patient cache database, where the patient cache database contains a large number of patient caches; - if no patient cache is available in the patient cache database for the patient of interest, - Entering the medical information about the patient of interest into the language model; - Generating, using the language model, a new patient cache based on the entered medical information, or - if a patient cache of interest to the patient is available in the patient cache database, - Retrieving the patient cache that corresponds to the patient of interest from the patient cache database, - Providing the received patient cache as context to the language model; - Entering the medical information about the patient of interest into the language model; - Update, using the language model, the existing patient cache for the patient of interest based on the entered medical information; - Updating the patient cache database with the newly created or updated patient cache.

[0095] In other words, an information source system or an information request system may detect that no patient cache is available for a patient of interest for whom medical information has been received. In this case, the medical information about the patient of interest can be entered into the language model—that is, the request language model or the source language model—to create a patient cache. The patient cache can then be stored in the patient cache database.

[0096] When it is detected that new medical information is available about a patient of interest for whom a patient cache already exists, the patient cache can be retrieved from the patient cache database and fed into the language model. The language model can then be instructed to update the patient cache based on the existing patient cache and the new medical information about the patient of interest. This ensures that the patient cache is always up to date. For example, whenever new data about a patient of interest is collected and stored in an information silo—that is, in the information request system or the information source system—this data can be incorporated into the patient cache by loading the patient-specific stored patient cache and additionally feeding the new data into the language model context.

[0097] Some aspects of the claimed procedures may be applicable to both the information request system and the information source system. Therefore, where the term "language model" is used, it should encompass the request language model and / or the source language model. Similarly, where the term "patient cache database" is used, it should encompass the requester-patient cache database and / or the source-patient cache database.

[0098] According to another aspect, a data processing device is provided which includes means for carrying out the steps of the method according to one of the claimed embodiments.

[0099] According to another aspect, a computer program product is provided, wherein the computer program product includes instructions which, when the program is executed by a computer, cause the computer to execute one of the embodiments mentioned above.

[0100] The computer program product, such as a computer program tool, can be implemented as a memory card, a USB flash drive, a CD-ROM, a DVD, or as a file that can be downloaded from a server on a network. Such a file can be delivered, for example, by transmitting the file containing the computer program product over a wireless communication network.

[0101] According to another aspect, a computer-readable storage medium is provided which includes instructions which, when executed by a computer, cause the computer to perform the steps of the procedure according to one of the embodiments mentioned above.

[0102] The invention also relates to a system that is specifically designed and / or configured for running and / or executing the computer-implemented method according to the invention. The system may include a patient cache database, a language model, a medical database, an authorization service, an automatic prompter interface, and / or a communication interface. The automatic prompter interface may include at least one processor.

[0103] The patient cache database can store a multitude of patient caches, each containing medical information about a patient. The language model can be trained to provide a natural language response to a given prompt. Specifically, the language model can be configured to generate a query message and / or a response message to a query message upon receiving a prompt, based on a patient cache corresponding to a patient of interest. The patient cache can be retrieved from the patient cache database and provided to the language model as context. The communication interface can be configured for transmitting, i.e., sending and / or receiving query messages and / or response messages.

[0104] The automatic prompter interface / processor can be designed to generate a prompt which, when the prompt is entered into the language model, causes the language model to - Generating, based on a patient cache fed as context to the language model, a query message with a prompt, the prompt being configured to be input into a source language model of at least one information source system to cause the source language model to provide missing medical information about a patient of interest; and / or - Identifying, based on a patient cache that is fed into the language model as context, missing medical information in the patient cache; and / or - Identify, based on a patient cache fed as context to the language model, at least one potential information source system that can provide the missing medical information; and / or - Creating a new patient cache from supplied medical data if no patient cache is available for a patient of interest in the patient cache database, or extending an existing patient cache for a patient of interest based on supplied medical data if a patient cache is available for the patient of interest in the patient cache database; and / or - Generating a response to a query message by entering the prompt into the language model, where the response indicates at least: - if the missing medical information is available in the at least one received patient cache, the missing medical information, or, - if the missing medical information is not available in the at least one patient cache obtained, indicate that the missing medical information is not available; and / or - Verify that the authentication information contained in a query message or a response message is correct.

[0105] Furthermore, the automatic prompter interface / processor may be designed to input the prompt into the speech model.

[0106] The embodiments and features described with reference to the method of the present invention apply mutatis mutandis to the device and / or system of the present invention.

[0107] Overall, the claimed methods and systems enable rapid information gathering from multiple sources within a network, without requiring predefined information protocols. This makes the claimed approach particularly advantageous when information needs to be exchanged between different information silos, such as between different institutions. Reducing the need for manual information requests leads to less distraction in the medical workflow and improved patient care.

[0108] Furthermore, possible implementations or alternative solutions of the invention also include combinations—not explicitly mentioned here—of features described above or below with regard to the embodiments. The person skilled in the art can also add individual or isolated aspects and features to the most basic form of the invention.

[0109] Further embodiments, features and advantages of the present invention will become apparent from the following description and the dependent claims, in conjunction with the accompanying drawings, in which the following applies: Fig. Figure 1 shows a schematic illustration of a procedure for generating a query message and for generating a response message to a query message; Fig. Figure 2 shows a schematic illustration of a procedure for generating another query message; Fig. Figure 3 shows a schematic illustration of another method for generating another query message; Fig. Figure 4 shows a schematic illustration of an embodiment of an information request system or an information source system; Fig. Figure 5 shows a schematic illustration of an embodiment of a health information network; Fig. Figure 6 shows a schematic illustration of another embodiment of a health information network; and Fig. Figure 7 shows a schematic illustration of an exemplary information flow in one embodiment of a health information network.

[0110] In the figures, identical reference numbers denote identical or functionally equivalent elements, unless otherwise specified. To improve the clarity of this text, some of the described examples are explained only in the context of generating a query message. However, the principles can be applied equally to generating a response message.

[0111] Fig. Figure 1 shows a schematic illustration of a method for generating a query message QM and for generating a response message RM to a query message QM. The method comprises several steps. The order of the steps does not necessarily correspond to the numbering of the steps, but may vary between different embodiments of the present invention. Furthermore, individual steps or a sequence of steps may be omitted and / or repeated.

[0112] Optionally, the process can be triggered by step S0, in which a user can request the generation of a query message (QM) to retrieve missing patient information. Additionally or alternatively, the request can be generated automatically, for example, at regular intervals or when new medical information (MI) about a patient of interest becomes available.

[0113] In step S10, a patient cache 22 is obtained. Patient cache 22 contains medical information (MI) of a patient of interest. However, the medical information (MI) and patient cache 22 may not be complete. This means that there may be at least one information gap in the patient cache and that medical information (MI) may be missing, which is not available in patient cache 22.

[0114] The patient cache 22 can contain a raw version of the available medical information (MI) about a patient. At least some of the medical information (MI) is preferentially preprocessed in the patient cache 22. Medical information (MI) that was available in images or PDF documents can, for example, be "translated" into textual information so that it is accessible to a language model and can be processed efficiently. In some examples, there may be too much available information about a patient to adapt the information to the context size of a language model 24. Therefore, at least some of the medical information (MI) in the patient cache 22 can result from generating a summary of medical information (MI) about the patient of interest. For example, lengthy medical reports can be summarized and key findings extracted.Additionally or as an alternative, features detected in radiological images can be described in text form.

[0115] In step S20, the patient cache 22 can be supplied as context to a request language model 24. There can be various ways to supply context to the request language model 24. For example, the patient cache 22 can be loaded from a patient cache database 20. In one example, the patient cache 22 could be a chat history representing a history of interactions between a user / automatic prompter interface 12 and a language model 24.

[0116] Optionally, in step S22, a prompt is entered into the request language model 24 to cause the request language model 24 to identify missing medical information (MI) in the patient cache 22. The request language model 24 can analyze the medical information (MI) contained in the patient cache 22 to identify which medical information (MI) is required or generally helpful and / or to solve a specific problem, such as making a specific diagnosis.

[0117] Optionally, in step S24, a prompt is entered into the request language model 24 to cause the request language model 24 to identify a potential information source system 10 from which the missing medical information (MI) can be obtained. For example, the request language model 24 can analyze the sources of the medical information (MI) present in the patient cache 22. The request language model 24 can identify sources that have previously provided information to the patient cache similar to the missing medical information (MI). These sources can be considered potential information source systems 10 for the missing medical information (MI).

[0118] In step S30, the request language model 24 is prompted, based on the patient cache 22 corresponding to the patient of interest, to generate a query message QM. The query message QM contains at least one prompt configured to be input into a source language model 24 of an information source system 10. When input into the source language model 24, the prompt causes the source language model 24 to provide the missing medical information MI about the patient of interest. Specifically, in response to receiving the prompt, the source language model 24 can generate a response message RM, which may contain the missing medical information MI.

[0119] In step S40, a source contact address of the at least one information source system 10 can be obtained. In one example, the source contact address is obtained from the medical information MI contained in the patient cache 22 by prompting the request language model 24 to search for a source contact address corresponding to the at least one identified potential information source system 10. The contact address may, for example, be contained in the header of a letter from a medical institution or may be derived from previous communications between the information source system 10 and the information request system 10.

[0120] The contact address used for communication between the information source system 10 and the information request system 10 is preferably implemented as an email address. This has the advantage that email addresses are widely available and easily searchable. Therefore, the claimed method and systems can be easily integrated into an existing information exchange structure.

[0121] In step S50, the generated query message QM is sent to at least one information source system 10.

[0122] In addition to the prompt, the QM query message can include at least one of the following features: - Information about the patient of interest, such as a unique patient identifier, the patient's name, the patient's date of birth, a patient ID, etc., - Authentication information that can be used by an information receiving system to verify the identity of the sender of a query message QM or a response message RM, - a unique query ID, and / or - a timeframe within which a response to the query message QM is expected.

[0123] In step S60, the query message QM, sent by the information request system 10, is received by an information source system 10. As explained above, the query message QM includes at least one prompt, which is configured to be input into a source language model 24 of the information source system 10. The prompt causes the source language model 24 to provide missing medical information (MI) about the patient of interest.

[0124] The query message QM can contain a certain tag or marker that identifies the query message QM as belonging to the communication framework of the invention and that allows the information source system 10 to distinguish the query message QM from other communications not belonging to the communication framework of the invention. For example, the query message QM can be an email addressed to the email address of the potential information source system 10. The marker can be a specific element of the email's subject line, e.g., a tag such as "FAXIQREQ" and / or a bottom line in the email body.

[0125] Optionally, in step S62, the authentication information contained in the query message QM is verified by the information source system 10. This ensures that the sender of the query message QM can be trusted and prevents misuse of the information provided by the information source system 10.

[0126] In one example, the query message QM includes a token issued by a centralized authentication system 14. This token can be verified by the information source system 10 to securely identify that the information request contained in the query message QM originates from a trusted source. Additionally, or alternatively, a hash, such as a numeric tag, can be included in the query message QM, where the hash refers to the query message QM or the response message RM previously exchanged between the information request system 10 and the information source system 10.

[0127] To enable the information request system 10 to verify that the received response message RM originates from a trusted potential information system 10, authentication information may also be included in response messages RM.

[0128] Optionally, in step S64, a current query ID is retrieved from the query message QM. For example, the unique query ID might be contained at a certain location within the query message QM, allowing it to be automatically extracted for further processing.

[0129] Optionally, in step S66, it can be checked whether the query message QM has already been answered. For example, this can be done by comparing the current query ID of the current query message QM with query IDs in a list of query IDs that correspond to queries previously answered by the information source system 10. If the information source system 10 determines that it has already answered the query message QM, the execution of the method is terminated. As a result, unnecessary processing of already answered information requests is avoided, and the operational efficiency of the method and the system according to the invention is improved.

[0130] In step S70, a source patient case database 20 is provided, wherein the source patient cache database 20 contains a multitude of patient caches 22. Each of the patient caches 22 contains a bundle of medical information (MI) about a patient. The source patient case database 20 is preferably stored within the premises of the information source system 10 to ensure the privacy of the stored medical patient information (MI).

[0131] In step S80, at least one patient cache 22 corresponding to the patient of interest is retrieved from the source patient cache database 20. The relevant patient cache 22 can be identified by comparing the patient identification information contained in the query message QM with the patient identification information in the stored patient caches 22 to identify the at least one relevant patient cache 22. The relevant patient cache 22 can then be loaded from the source patient cache database 22.

[0132] In step S90, the obtained patient cache 22 can be fed to the source language model 24 to be available as context for the source language model 24.

[0133] In step S100, a response message RM to the query message QM is generated by entering the prompt of the query message QM into the source language model 24. The generated response indicates, at a minimum, whether the requested missing medical information MI is available in the information source system 10. If the information is available, the requested missing medical information MI is provided. To locate the missing medical information MI, the source language model 24 can search the received patient cache 24 for the missing information MI specified in the query message QM.

[0134] In step S110, a requester contact address is obtained, which is required to send the response message RM to the information request system 10. The requester contact address is preferably included in the query message QM and is obtained simply by replying to the requester contact address specified in the query message QM. Additionally, or as an alternative, the requester contact address can be identified by entering an appropriate prompt into the source language model 24 to identify a potential requester contact address from the query message QM and / or the received patient cache 22.

[0135] In step S120, the generated response message RM is sent to the information request system 10 using the requester contact address.

[0136] Communication between the information request system 10 and the information source system 10 can preferably be encrypted, e.g., using TLS-encrypted emails, to ensure the privacy of the transmitted medical information (MI).

[0137] There may be instances where the requested medical information (MI) is not available at the queried information source system 10. In such cases, the requesting information request system 10 and / or the queried information source system 10 can identify and query other potential sources of information, as described in Fig. 2 and Fig. 3 is visualized.

[0138] Fig. Figure 2 visualizes a schematic illustration of a procedure for generating another query message QM in an information requirements system 10.

[0139] At step S200, a response message RM is received from information source system 10. The response message RM indicates that the missing medical information MI is not available in information source system 10.

[0140] In step S210, the request language model 24 can be prompted to identify at least one additional potential information source system 10 that can provide the missing medical information (MI). For example, the prompt fed to the request language model 24 may have different wording than the prompt provided in the first iteration. Alternatively, the criteria for determining that a search result is classified as a potential information source 10 may be relaxed; for example, a threshold for a result relevance parameter may be lowered.

[0141] In step S220, a further query message QM can be generated using the request language model 24, wherein the further query message QM includes at least one prompt that is set up to be entered into a source language model 24 of the further information source system 10, to cause the source language model 24 to provide the missing medical information MI.

[0142] In step S230, an information source contact address of the further potential information source system 10 is obtained.

[0143] In step 240, the generated query message QM is sent to the further information source system 10 using the information source contact address.

[0144] Therefore, the procedure after step S210 can, in principle, be executed similarly to the "standard" procedure for generating a QM query message. Obviously, features or steps of the various procedures for generating a QM query message are applied to other embodiments of the procedure or integrated into other embodiments of the procedure.

[0145] Fig. Figure 3 visualizes a schematic illustration of a procedure for generating another query message QM in an information source system 10. According to this procedure, the information source system 10 of the first iteration becomes the information request system 10 of the second iteration. This allows for a rapid scan of an entire network of multiple information systems 10 for missing medical information MI.

[0146] In step S310, at least one additional potential information source system 10 is identified that can provide the missing medical information (MI). To do this, the source language model 24 is prompted accordingly, and the received patient cache 22 and / or the query message QM are analyzed by the source language model 24 to find further sources from which the missing information (MI) can be retrieved. Since the "knowledge" contained in the patient cache 22 of the source system is most likely to differ from the "knowledge" in the patient cache 22 of the information request system, there is a probability that a promising additional source for missing medical information (MI) can be found by searching the patient cache 22 of the information source system for further sources.

[0147] From there, the procedural steps, as known from the information requirements system 10 and the procedure for generating a query message QM, can be carried out.

[0148] In step S320, a query message QM is generated using the source language model 24, wherein the query message QM includes a prompt that is set up to be entered into a source language model 24 by at least one other information source system 10, to cause the source language model 24 to provide the missing medical information MI about the patient of interest.

[0149] In step S330, an information source contact address of at least one other information source system 10 is obtained.

[0150] In step S340, the generated query message QM is sent to at least one other information source system 10 using the information source contact address.

[0151] Fig. Figure 4 shows a schematic illustration of an embodiment of a health information system 10, i.e. an information request system 10 or an information source system 10.

[0152] The health information system 10 contains medical information (MI) about various patients. For example, medical images can be stored in a PACS system within the health information system 10. Furthermore, radiological reports and / or EMRs can be stored within the health information system 10, e.g., in a RIS or on other storage systems. Together, these storage systems for medical information (MI) and / or medical data constitute a medical information database 26.

[0153] It is a common problem in the healthcare sector that medical information (MI) available within a health information system (HIS), i.e., in the health information database (HIV), is difficult to access or even search from outside the HIS. This means that the HIS forms an information silo, and the exchange of information between different silos is difficult.

[0154] To enable and facilitate the exchange of information between multiple information silos, the health information system 10 according to the invention includes an automatic prompter interface 12 for managing the information flows to and from the health information system 10. The automatic prompter interface 12 is designed to input prompts into a language model 24 of the health information system 10 in order to cause the language model 24 to perform various tasks, e.g., identifying missing medical information (MI) and generating query messages (QM) to retrieve missing medical information (MI).

[0155] To make the available medical data / medical information (MI) more accessible for language model 24 and to enable faster processing of the information, the medical information (MI) is grouped into patient caches 22, with each patient cache 22 corresponding to a specific patient. Therefore, each patient cache 22 contains a bundle of medical information (MI) relating to that specific patient.

[0156] The patient caches 22 can be stored in the patient cache database 20. Thus, after creation, the patient caches 22 can be easily loaded from and / or stored in the patient cache database 20 without having to create a patient cache 22 each time the medical information (MI) available about a patient is examined. In an example, the patient caches 22 represent a specific state of the language model 24. Therefore, by storing a patient cache 22 in the patient cache database 20, the current state information of the language model 24 is stored. The information state can be restored by loading the patient cache 22 from the patient cache database 20.

[0157] The health information system 10 also includes an authentication system 14, which may be part of a centralized authentication system 14, as explained previously, or which may not be centralized, i.e., specific to the health information system 10.

[0158] Furthermore, the health information system 10 can include a unique query ID generator 16 for generating query IDs to be included in the query message QM and / or the response messages RM sent by the health information system 10.

[0159] The health information system 10 may include a communication interface 18 for transmitting, i.e. sending and / or receiving, query messages QM and response messages RM.

[0160] Fig. Figure 5 shows a schematic illustration of an embodiment of a health information network 100. The health information network 100 includes a plurality of health information systems 10. Depending on the operating conditions, at least one of the health information systems 10 can serve as an information request system 10, and at least one of the other health information systems 10 can serve as an information source system 10.

[0161] The health information systems mentioned above can be interconnected via a communication network N. Examples of communication networks N include private or public local area networks (LANs), wireless local area networks (WLANs), metropolitan area networks (MANs), wide area networks (WANs), the internet, and / or combinations thereof. The communication network N can involve wired and / or wireless communications according to one or more standards and / or over one or more transport media. Communication over the network N can be conducted according to a wide variety of communication protocols, such as Transmission Control Protocol and Internet Protocol (TCP / IP), User Datagram Protocol (UDP), and IEEE communication protocols.In one example, network N can include wireless communications according to Bluetooth specification sets or another standard or proprietary wireless communication protocol. In another example, network N can also include communications over a cell-based network, including, for example, a GSM (Global System for Mobile Communications), CDMA (Code Division Multiple Access), or EDGE (Enhanced Data for Global Evolution) network.

[0162] The connections between the components within a health information system 10, i.e. between the automatic prompter, the language model, the patient cache database, etc., can be implemented similarly.

[0163] The health information systems 10 communicate with each other via their communication interfaces 18, so that the query messages QM are sent via the network N and the response messages RM are sent as a reply.

[0164] Fig. Figure 6 shows a schematic illustration of another embodiment of a health information network 100. The illustrated embodiment of the health information network 100 has a large number of common features with the health information network 100. To avoid repetition, these common features are referred to in connection with Fig. 6 is not described again. Instead, the explanations associated with Fig. 5 were made, applicable.

[0165] The health information network 100 comprises a single language model 24 that communicates with some or all of the health information systems 10. The language model 24 can be accessed via the network N; for example, medical information MI, patient caches 22, and / or prompts are exchanged between the central language model 24 and the decentralized automatic prompter interfaces 12.

[0166] Communication between the automatic prompter interfaces 12 of the health information systems 10 and the language model 24 can be encrypted to ensure data privacy. For example, the language model 24 can be hosted by / stored on the premises of a trusted institution, guaranteeing that medical data of patients fed into the language model 24, e.g., in the form of patient caches 22, is treated confidentially.

[0167] Thus, the main advantage of using a shared language model 24, which is shared between multiple health information systems 10, is that the language model can be larger and capable of generating better responses than smaller, decentralized models. With a shared language model 24, sufficient storage capacity does not need to be provided to store a large-scale language model in each individual health information system 10. Instead, a shared language model 24 can be hosted centrally to save storage space.

[0168] Fig. Figure 7 shows a schematic illustration of an exemplary information flow in an embodiment of a health information network 100.

[0169] The health information network 100 comprises four health information systems 10, specifically a first health information system 10.1, a second health information system 10.2, a third health information system 10.3, and a fourth health information system 10.4, which communicate with each other. In the present example, the first health information system 10.1 is considered the information request system. Missing medical information (MI) about a patient of interest is identified in the first health information system 10.1. A query message (QM) is generated in the first health information system 10.1, for example, according to one of the procedures described above. The query message (QM) is sent to the second health information system 10.2, which has been identified as a potential source for the missing medical information (MI).

[0170] However, after entering the prompt contained in the query message QM into the source language model 24 of the second health information system 10.2, it is detected that the missing medical information MI is not available in the second health information system 10.2. Therefore, a corresponding response message RM is generated and sent back to the first health information system 10.1. The response message RM indicates that the missing information is not available in the second health information system 10.2. The response message RM further indicates that the second health information system 10.2 has identified other potential information source systems, namely the third information source system 10.3 and the fourth information source system 10.4, from which the missing medical information MI can be retrieved.

[0171] As a result, another query message QM is generated in the second health information system 10.2 and the further query message QM is sent to the third health information system 10.3 and to the fourth health information system 10.4.

[0172] The missing medical information MI is not available in the third health information system 10.3, so a corresponding (negative) response message RM is generated and sent to the second health information system 10.2. However, the third health information system 10.3 also identifies the fourth health information system 10.4 as a potential source for the missing information, generates a corresponding query message QM, and sends it to the fourth health information system 10.4.

[0173] Thus, the fourth health information system, 10.4, receives the query message QM from the second health information system, 10.2. Searching the patient cache corresponding to the patient of interest, it is detected that the missing medical information MI is available in the fourth health information system, 10.4. A response message RM containing the missing medical information MI is generated and sent to the second health information system, 10.2, which forwards the response message RM to the first health information system, 10.1. Alternatively or additionally, the response message RM containing the missing medical information MI can be sent directly to the original information request system, i.e., the first health information system, 10.1, as indicated by the dashed line.

[0174] Since the missing medical information (MI) has already been delivered to the first health information system (10.1), responding to the query message (QM) sent from the third health information system (10.3) to the fourth health information system (10.4) is unnecessary. This is detected in the fourth health information system (10.4) because the unique query ID contained in the query message (QM) is already known to the fourth health information system (10.4). The fourth health information system (10.4) can therefore determine that the query has already been answered and that the process can be terminated without generating a further response message (RM) to the query message (QM) received from the third health information system (10.3).Thus, the entire health information system can be scanned for missing medical information (MI) 100, while the process remains efficient as unnecessary calculations are avoided.

[0175] In summary, the methods and systems according to the invention allow for the rapid collection of medical information from a wide variety of sources within a health information network. Manual activities in the information collection process are drastically reduced, thus effectively preventing distractions for medical professionals. In some examples, information can be retrieved even before it is requested by those skilled in the art, so that the information is immediately available when it is of interest to them. This improves access to medical information and can accelerate the clinical decision-making process.

[0176] The various illustrative logic blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above in general terms with regard to their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and design constraints imposed on the overall system.The average person skilled in the art can implement the described functionality in varying ways for each specific application; however, such implementation decisions should not be interpreted as causing a deviation from the scope of protection of this disclosure or the claims. Embodiments implemented in computer software may be implemented in software, firmware, middleware, microcode, hardware description languages, or any combination thereof. A code segment or machine-executable instructions may represent a procedure, a function, a subroutine, a program, a routine, a subroutine, a module, a software package, a class, or any combination of instructions, data structures, or program instructions.A code segment can be coupled to another code segment or a hardware circuit by passing and / or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc., can be passed, forwarded, or transferred by any suitable means, including memory release, message forwarding, token forwarding, network transmission, etc.

[0177] The actual software code or specialized control hardware used to implement these systems and methods does not limit the claimed features or this disclosure. Thus, the operation and behavior of the systems and methods have been described without reference to the specific software code, which implies that software and control hardware can be designed to implement the systems and methods based on the description herein.

[0178] When implemented in software, the functions can be stored as one or more instructions or code on a non-transitory computer-readable or processor-readable storage medium. The steps of a method or algorithm disclosed herein can be implemented in a processor-executable software module, which may reside on a computer-readable or processor-readable storage medium. A non-transitory computer-readable or processor-readable medium includes both computer storage media and tangible storage media that facilitate the transfer of a computer program from one location to another. A non-transitory processor-readable storage medium can be any available medium accessible by a computer.By way of example, and not limitation, such non-transitory processor-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other tangible storage medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer or processor. Disk and disc, as used here, include Compact Disc (CD), Laser Disc, Optical Disc, Digital Versatile Disc (DVD), Floppy Disc, and Blu-ray Disc, with disks typically reproducing data magnetically, whereas discs reproducing data optically using lasers. Combinations of the above should also be included in the scope of computer-readable media.In addition, the operations of a procedure or algorithm can reside as one or any combination or set of codes and / or instructions on a non-transitory processor-readable and / or computer-readable medium, which may be integrated into a computer program product.

[0179] The preceding description of the disclosed embodiments is intended to enable the person skilled in the art to implement or use the embodiments and variations thereof described herein. Various modifications to these embodiments are readily apparent to the person skilled in the art, and the principles defined herein can be applied to other embodiments without altering the nature or scope of protection of the invention disclosed herein. Thus, the present disclosure is not intended to be limited to the embodiments shown herein, but rather to grant the broadest possible scope of protection in accordance with the following claims and the principles and novel features disclosed herein.

[0180] Although various aspects and embodiments have been disclosed, other aspects and embodiments are considered. The various aspects and embodiments disclosed herein serve only as illustrations and are not intended to be limiting, the true scope and nature of which are specified by the following claims.

Claims

[1] A computer-implemented method for generating, in an information request system, a query message (QM) with a prompt, wherein the prompt is configured to be input into a source language model of an information source system separate from the information request system, in order to cause the source language model to provide missing medical information (MI), wherein the method comprises the following steps: - Receiving (S10) a patient cache (22) relating to a patient of interest, wherein the patient cache (22) includes medical information (MI) about the patient of interest; - Providing (S20) the received patient cache (22) as context to a demand language model; - Generate (S30), using the request language model and based on the patient cache (22), a query message (QM) with a prompt, wherein the prompt is configured to be input into a source language model of at least one information source system to cause the source language model to provide missing medical information (MI) about the patient of interest; - Obtain (S40) an information source contact address of at least one information source system; - Sending (S50) the generated query message (QM) to the at least one information source system using the information source contact address. [2] The method of claim 1, wherein the method further comprises the following step: - Identify (S22), using the request language model and based on the patient cache (22), missing medical information (MI) in the patient cache (22); and / or - Identify (S24), using the request language model and based on the patient cache (22), at least one potential information source system that can provide the missing medical information (MI). [3] Method according to any of the preceding claims, wherein the step of obtaining (S40) an information source contact address includes identifying the information source contact address of the potential information source system with the request language model based on the obtained patient cache (22). [4] A method according to any of the preceding claims, the method further comprising the following steps: - Receive (S200) a response from the information source system indicating that the missing medical information (MI) is not available in the information source system; - Identify (S210), using the request language model and based on the patient cache (22), at least one other potential information source system that can provide the missing medical information (MI); - Generate (S220), using the request language model and based on the patient cache (22), at least one further query message (QM) with a prompt, wherein the prompt is configured to be input into a source language model of at least one further information source system to cause the source language model to provide the missing medical information (MI) about the patient of interest; - Receive (S230) an information source contact address of the wider information source system; - Sending (S240) the generated further query message (QM) to the further information source system using the information source contact address. [5] Method according to any of the preceding claims, wherein the method is carried out at regular intervals, on request and / or when new medical information (MI) about a patient of interest is available. [6] A computer-implemented method for generating, in an information source system, a response message (RM) to a query message (QM) from an information request system separate from the information source system, wherein the method comprises the following steps: - Received (S60) in the information source system, a query message (QM) from an information request system, wherein the query message (QM) includes at least one prompt, the prompt being configured to be input into a source language model of the information source system to cause the source language model to provide missing medical information (MI) about a patient of interest; - Providing (S70) a source patient cache database containing a multitude of patient caches (22), each patient cache (22) containing medical information (MI) about a patient; - Received (S80), from the source patient cache database, at least one patient cache (22) relating to the patient of interest; - Providing (S90) the received patient cache (22) as context for the source language model of the information source system; - Generating (S100) a response message (RM) to the query message (QM) by entering the prompt into the source language model, where the response message (RM) at least specifies: - if the missing medical information (MI) can be provided based on the at least one patient cache obtained (22), the missing medical information (MI), or, - if the missing medical information (MI) cannot be provided based on the at least one patient cache received (22), that the missing medical information (MI) is not available in the information source system; - Received (S110) a requester contact address from the information request system; - Send (S120) the generated response message (RM) to the information request system using the requester contact address. [7] The method of claim 6, wherein the missing medical information (MI) is not available in the information source system and wherein the method further comprises the following steps: - Identify (S310), using the source language model and based on the received patient cache (22) and / or based on the query message (QM), at least one other potential information source system that may be able to provide the missing medical information (MI); - Generate (S320) a query message (QM) using the source language model with a prompt configured to be input into a source language model of at least one other information source system to cause the source language model to provide missing medical information (MI) about the patient of interest; - Obtain (S330) an information source contact address of at least one other information source system; - Sending (S340) the generated query message (QM) to at least one other information source system using the information source contact address. [8] Method according to one of claims 6 or 7, wherein the query message (QM) includes a unique query ID, and the method further comprises the following steps: - Retrieve (S64) a current query ID from the query message (QM) to which a response should be made; - Check (S66) whether the query message (QM) has already been answered by comparing the current query ID with query IDs in a query ID list, where the query ID list contains query IDs of query messages (QM) that have previously been answered by the information source system; - if the query message (QM) has already been answered, terminate the execution of the procedure. [9] Method according to any of the preceding claims, wherein the query message (QM) and / or the response message (RM) further include authentication information, wherein the authentication information is designed to be verified in the information request system or in the information source system. [10] Method according to any of the preceding claims, wherein the requester contact address and / or the information source contact address is an email address. [11] Method according to any of the preceding claims, wherein the query message (QM), the response message (RM) and / or the patient cache (22) further include a patient identifier relating to the patient of interest. [12] A method according to any of the preceding claims, the method further comprising the following steps: - Obtaining medical information (MI) about a patient of interest; - Check whether a patient cache (22) is available for the patient of interest in a patient cache database (20), wherein the patient cache database (20) contains a plurality of patient caches (22); - if no patient cache (22) of interest is available in the patient cache database (20), - Entering the medical information (MI) about the patient of interest into the language model (24); - Generating, using the language model (24), a new patient cache (22) based on the entered medical information (MI), or - if a patient cache (22) of interest to the patient is available in the patient cache database (20), - Retrieving the patient cache (22) relating to the patient of interest from the patient cache database (20), - Providing the received patient cache (22) as context to the language model (24); - Entering the medical information (MI) about the patient of interest into the language model (24); - Update, using the language model (24), the existing patient cache (22) for the patient of interest based on the entered medical information (MI); - Updating the patient cache database (20) with the newly created or updated patient cache (22). [13] Data processing equipment comprising means for carrying out the steps of the method according to any one of claims 1 to 12. [14] Computer program product comprising instructions which, when the program is executed by a computer, cause the computer to perform the steps of the method according to any one of claims 1 to 12. [15] Computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the steps of the method according to any one of claims 1 to 12.

Images