Modular clinical decision support system for diagnosis and treatment

Abstract

A decision support system comprises a decision tree database (14) and a path history database (15) which are stored on a controlling server (16) that is connected via a network (17) to a remote device (18). The decision tree database (14) contains specialized modules (10) and intermodular links (11). The modules (10) contain decision trees (1) of related clinical disciplines (13) and intramodular links (25). The path history database (15) contains path data for all completed sessions (29). The remote device (18) contains an interface (19) that includes: a raw data selection menu (20), a step display and selection window (21) and a path history (22).

Description

[0001] MODULAR CLINICAL DECISION SUPPORT SYSTEM FOR DIAGNOSTICS AND TREATMENT

[0002] Description

[0003] Field of invention

[0004] The invention relates to “knowledge” systems for supporting medical decision-making based on a decision tree.

[0005] Prerequisites for the creation of an invention

[0006] There are two types of clinical decision support systems (CDSS) used in diagnosis and treatment:

[0007] 1. “Unknowledgeable” SPPVRs, using various artificial intelligence architectures based not on the conclusions of scientific research, but, for example, on the results of processing collected statistical data using mathematical methods,

[0008] 2. “Knowledge” SPPVR, based on scientific knowledge.

[0009] Regardless of the type of SPVR, all of them have a number of disadvantages when used in real clinical practice. These disadvantages are described in detail in the following sources and are outlined below:

[0010] 1. Reed T. Sutton et Al., An overview of clinical decision support systems: benefits, risks, and strategies for success, NPJ Digit Med. 2020; 3:17,

[0011] 2. Bangui Khan et Al., Drawbacks of artificial intelligence and their potential solutions in the healthcare sector, Biomed Mater Devices. 2023 Feb 8: 1-8

[0012] 3. Marsa Gholamzadeh et Al., The application of knowledge-based clinical decision support systems to enhance adherence to evidence-based medicine in chronic disease, J Healthc Eng. 2023; 2023: 8550905.

[0013] All disadvantages can be divided into 3 groups:

[0014] 1. Common disadvantages of all PPVRS: a. Difficulty of integration into the clinical process - the clinical workflow is complex and clinicians do not have enough time to work in parallel with the PPVRS, which makes it difficult to use the PPVRS in real clinical practice.

[0015] 2. Disadvantages of “knowledgeless” IPS: a. “black box” problem – IPS cannot explain the reasons for the conclusions reached, as a result most clinicians do not use them directly for diagnosis due to uncertainty about the accuracy and reliability of the results, b. data collection problem – since AI-based IPS raises concerns about data security and privacy among institutions, there is a natural reluctance to share medical data, c. difficulty in analyzing data quality – some predictive algorithms may not be as successful in predicting future outcomes since they recreate the past, and medical records are rarely neatly organized since they often contain errors and are stored inconsistently, d. system dependence on database size – the less extensive the database, the lower the efficiency of the system’s training and the higher the probability of erroneous conclusions, e.Accountability issue - because clinical decision making is done in the SPPVR, it is not clear who is legally responsible if decisions are made incorrectly.

[0016] 3. Disadvantages of "knowledge" DSPS: a. contradictory data - sometimes different scientific papers contain contradictory information, which significantly complicates their use in DSPS. Given the above, knowledge-based DSPS remains the most preferable. They use a variety of architectures, among which the decision tree is the most accurate. Despite this, DSPS based on a decision tree has disadvantages:

[0017] 1. "sequential structure" - the process begins with a data entry node, followed by decision nodes, ending with a recommendation node - in real clinical practice, the process of diagnosis and treatment is not a "sequential" but a "parallel" process, in which treatment is adjusted as the diagnosis is refined.

[0018] 2. Multiple data entry point - a data entry node, which is an entry point, requires the aggregation of multiple patient data (complaints, anamnestic data, symptoms and clinical signs), entered manually or transferred from a database, the required completeness of which is often not available to the clinician at the initial stages of patient care in real clinical practice.

[0019] US 7577573 B2 9 / 2009

[0020] WO 2010 / 119356 A2 4 / 2009

[0021] WO 2014 / 037872 A2 3 / 2014

[0022] Summary of the invention

[0023] It would be useful to eliminate the shortcomings of the SPVR in general and the “knowledge” SPVR in particular.

[0024] In order to better solve the problem of "sequential structure", a first aspect of the invention proposes a system comprising: a decision tree constructed according to the alternating principle of the arrangement of nodes, in which the initial data node leads to a path, with contextually alternating decision nodes, additional data nodes, preliminary recommendation nodes, notification and alarm nodes, which end with a final recommendation node.

[0025] In order to better solve the problem of the difficulty of integration into the clinical process, the second aspect of the invention proposes a system comprising: specialized modules, each of which covers several related clinical disciplines, created on the basis of clinical guidelines and protocols, consisting of corresponding decision trees; and decision trees that can be connected to decision trees in one module and to decision trees in other modules, thereby creating a continuous process; and paths starting from the initial data, passing through certain branches of one or more decision trees in one or more modules and ending with final recommendations; and parameter calculators integrated into decision trees and implemented in five types, which serve to reduce the duration of the paths

[0026] Thus, the system covers the entire period of patient care, is integrated into the clinician’s work and accelerates the decision-making process.

[0027] In order to better solve the problem from the entry point with a lot of data, a third aspect of the invention proposes a system comprising: a decision tree starting from a source data node that requests only one parameter as the initial data about a patient (chief complaint, symptom, clinical sign or marker anamnestic information selected from a menu), whereby the work with the system can be started with limited information about the patient, and the system only requests additional data when necessary through context-dependent additional data nodes.

[0028] In order to better address the problem of data inconsistency, the fourth aspect of the invention proposes a solution consisting of: creating decision trees based only on clinical guidelines, protocols, recommendations and standards used taking into account the hierarchy of ranks and the evidence base of the sources.

[0029] Drawing Descriptions

[0030] Aspects of the invention are explained in the following drawings. The drawings are schematic and may not be drawn to scale. In the drawings, similar elements are designated by the same reference numbers.

[0031] Figure 1 is an example of a decision tree.

[0032] Figure 2 shows the graph structure of the above decision tree example.

[0033] Drawing 3 shows the general outline of the invention.

[0034] Drawing 4 shows the layout of the modules.

[0035] Figure 5 shows an example of a path route on a module diagram.

[0036] Figure 6 is a graph structure of the above path route. Figure 7 shows the transition from the primary data selection menu to the decision tree. Figure 8 shows a fragment of the interface with the primary data selection menu.

[0037] Figures 9A-B show a fragment of the interface with a display window and step selection. Figures 10A-E show examples of the graph structure of five types of parameter calculators. Figure 11 shows the hierarchy order of the sources used.

[0038] Figure 12 shows the order of the hierarchy of evidence of sources.

[0039] Detailed description of the invention

[0040] The first aspect of the invention

[0041] Figure 1 shows a decision tree 1 using the example path "abdominal pain" from the "emergency medical care" module. The decision tree has a primary data node 2, which is the starting point on a path through the decision tree. The tree also contains additional data nodes 6, which require additional data, condition nodes 4, which are the nodes with which the condition is associated, primary recommendation nodes 3, which define the current recommendations, notification nodes 9, which contain notifications and alarms that may be located throughout the decision tree and interleaved in a context-sensitive manner. Branches of the decision tree may terminate in leaf recommendation nodes 7, which are the ends of paths in the decision trees. Branches of the decision tree may terminate in transition nodes to other decision trees 5. Edges 8 connect different types of nodes to indicate possible routes along the nodes that the path may follow.The decision tree can be modeled as a graph structure and stored in a decision tree database 14.

[0042] Figure 2 shows the graph structure of the above example decision tree, indicating the primary data node 2, the additional data node 6, the condition node 4, the primary recommendation node 3, the notification node 9, the final recommendation node 7, the transition to other decision trees node 5, and the edges connecting the nodes 8.

[0043] This aspect of the invention allows, in parallel with the diagnostic stages, to provide recommendations for patient management corresponding to each diagnostic stage. The second aspect of the invention

[0044] Figure 3 shows a general diagram of the invention, including a decision tree database 14, a path history database 15, a control server 16, a network 17 and a remote device 18. The decision tree database 14 contains specialized modules 10, including decision trees of related disciplines 13. For example, the "EMERGENCY" module contains decision trees created on the basis of the "Emergency Medical Care" and "Disaster Medicine" protocols, and the "CSM" module contains "Resuscitation and Intensive Therapy", "Anesthesiology" and "Algology", respectively. The data is stored on the control server 16 and is connected via the network 17 to the remote device 18 (desktop computer, laptop, tablet, smartphone). The remote device 18 has an interface 19. The interface 19 contains access to the primary data selection menu 20, which is an intuitive menu for the physician to select leading complaints, symptoms, clinical signs, manifestations of anamnestic signs, incidents, known diagnoses and clinical situations.After selection, the user is redirected to the interactive window for displaying and selecting a step 21, through which the user, without entering text, simply clicks on the button to select conditions, data, and notes the recommendations that have been completed and the notifications that have been read. The interface 19 also contains the history of the paths 22, which displays in text format in chronological order all the stages of all completed sessions for each patient.

[0045] Figure 4 shows a diagram of the modules and their interrelations using the example of fragments of the modules ICU (intensive care module) 23, EMERGENCY (emergency care module) 24 and INTERDISCIPLINARY (interdisciplinary module) 33, including the modules themselves 10, decision trees 1, intra-module connections 25, examples of inter-module connections 11 (for simplicity, not all are shown), the primary data selection menu, which is the starting point of all paths 20, the end point of all sessions 26, single abbreviations, which are the name of the main decision tree (for example, CAR - cardiac arrest) 27, double abbreviations, which are the name of the auxiliary decision tree (for example, CAR PCA - cardiac arrest - care after cardiac arrest) 28.

[0046] In Figure 5, the thick grey line shows an example of a path 29 that starts in the primary data selection menu 20, passes through several decision trees 1 in modules 23, 24 and 33 and ends at the end point of all paths 26. In the example shown in the figure, the path uses decision trees sequentially:

[0047] CAR cardiac arrest (ambulance module)

[0048] CAR RSA care after cardiac arrest (emergency module)

[0049] CAR GSH general shock and hypotension after cardiac arrest (EMC module) CAR RAM patient comfort after cardiac arrest (EMC module) CAR AIM post-cardiac arrest airway management (EMC module) CAR BIA post-cardiac arrest placement (EMC module) CAR DOC post-cardiac arrest documentation (EMC module) CAS CCS cause of cardiogenic shock (Critical Care Module)

[0050] CAS MED drug treatment of cardiogenic shock (intensive care module) CAS HEM hemodynamic management in cardiogenic shock (intensive care module) CAS IAB IABP in cardiogenic shock (intensive care module)

[0051] CAS COR coronary revascularization in cardiogenic shock (intensive care module) CAS SPS specific situations in cardiogenic shock (intensive care module) CAS RES stabilization after cardiogenic shock (intensive care module) CAS DOC documentation after cardiogenic shock (intensive care module)

[0052] SRSP MMnST, with revascularization, shock and after arrest (interdisciplinary module) IAW SRSP weaning from IABP after MMnST, revascularization, shock, heart (interdisciplinary module) VAW SRSP weaning from mechanical ventilation after MMnST, revascularization, shock, arrest (interdisciplinary module)

[0053] REH SRSP rehabilitation after MMnST, revascularization, cardiogenic shock, cardiac arrest (interdisciplinary module)

[0054] CON SRSP control after MMnST, revascularization, cardiogenic shock, cardiac arrest (interdisciplinary module)

[0055] DIS SRSP discharge after revascularization MMnST, revascularization, cardiogenic shock, cardiac arrest (interdisciplinary module)

[0056] DOC SRSP documentation after revascularization MMnST, revascularization, cardiogenic shock, cardiac arrest (interdisciplinary module)

[0057] Figure 6 shows an example of the graph structure of the CAR PCA decision tree, with path 29 passing through branches of the decision trees highlighted in black. To eliminate the “loop effect,” in which the route of path 29 may cross the same node of the decision tree two or more times, the decision trees may contain one or more repeating fragments 31, marked in white on the diagrams.

[0058] All completed steps of all path routes 29 are stored in the path history database 14 and displayed in the path history window 22 of the interface 19, which eliminates the need for the clinician to document all diagnostic and treatment procedures, significantly reducing the bureaucratic burden on staff.

[0059] This aspect allows the system to cover all stages of patient care, self-correct depending on changes in the clinical situation, reduce bureaucratic burden and thereby integrate into the work of the clinician at all stages of patient care.

[0060] The third aspect of the invention

[0061] Figure 7 shows the menu for selecting primary data 20, containing data (chief complaints, symptoms, clinical signs, manifesting anamnestic signs, known diagnoses, clinical situations, incidents) 30. The figure also shows modules 10, containing decision trees 1, displaying primary data nodes 2. The figure also shows inter-module connections 11.

[0062] Each data 30 in the primary data selection menu 20 is linked to only one primary data node 2 of a particular decision tree 1. Each primary data node 2 is linked to one or more data 30 in the primary data selection menu 20.

[0063] Figure 8 shows a fragment of the interface with a menu for selecting primary data 20, consisting of a general menu containing a plurality of submenus, each of which contains a corresponding list of data 30. The sections and subsections are arranged in an order that is intuitively understandable for the clinician.

[0064] Figures 9A-B show a fragment of the interface with a window for displaying and selecting steps 21, with an example of an additional data node 6 that can appear during the session and refine the data as needed.

[0065] To reduce the length of paths, some additional data nodes 6 and condition nodes 4 are implemented in the form of parameter calculators of five types, in which, depending on the user's choice of parameters, the further route of the path is determined.

[0066] Figure 10A shows an example of the graph structure of a calculator of the first type with the possibility of combining parameters.

[0067] Figure 10B shows an example of the graph structure of a calculator of the second type with the ability to sum parameters. Figure 10C shows an example of the graph structure of a calculator of the third type with the ability to simply select one parameter from several.

[0068] Figure 10 D shows an example of the graph structure of a type 4 calculator with the ability to sum parameters in combination with the selection of an additional key parameter 32.

[0069] Figure 10E shows an example of the graph structure of a calculator of the fifth type with the possibility of summing parameters selected from individual parameter groups 33.

[0070] This aspect allows the clinician to begin working with a patient by selecting just one symptom or sign from a list of data, then the system requests additional data as needed, eliminating the need to waste time collecting large amounts of data that are not needed in a particular case.

[0071] The fourth aspect of the invention

[0072] Figure 11 shows in the form of pyramids: the hierarchy of ranks of the sources used (protocols, guidelines, recommendations and standards) 34, in which the closer to the top, the more preferable the source, and the hierarchy of the evidence base of sources 35, in which the closer to the top, the more preferable the evidence base.

[0073] In case of discrepancy between data in different sources of different ranks, preference is given to the source of the higher rank.

[0074] In case of discrepancies in data in different sources of the same rank related to gender, age and population factors, these differences are taken into account when constructing decision trees.

[0075] In case of discrepancy between data in different sources of the same rank, not related to gender, age and population factors, a comparison of the evidence base of each source is carried out.

[0076] This aspect of the invention eliminates the problem of data inconsistency.

Claims

Formula:

1. A clinical decision support system, characterized in that it contains a control server (16) storing a decision tree database (14) with interconnected specialized modules (10) and a path history database (15), which is connected to a remote device (18) via a network (17) and a control server (16).

2. The system according to claim 1, characterized in that the database of decision trees (14) contains specialized modules (10) connected to each other by intermodule links (11), consisting of decision trees (1) and intramodule links (25), which together form path routes (29).

3. The system according to item 2, characterized in that the decision trees (1), along with the primary data nodes (2) and the final recommendation nodes (7), contain additional data nodes (6), condition nodes (4), primary recommendation nodes (3), notification nodes (9), nodes for transition to other decision trees (5) and duplicate fragments of other decision trees (31).

4. The system according to claim 1, characterized in that the database of route histories (15) contains data on the routes of all completed routes (29) in chronological order.

5. The system according to item 1, characterized in that the user interface (19) of the remote device (18) contains a menu for selecting primary data (20), with the possibility of selecting individual data (30), without selecting a combination of data, with a transition to the display window and selection of steps (21).

6. The system according to claim 1, characterized in that the user interface (19) of the remote device (18) contains a path history window (22) displaying all stages of all previously completed path routes (29) in chronological order.

7. The method for creating decision trees (1) according to paragraph 3, taking into account the hierarchy of source ranks (34) and the hierarchy of the evidence base (35).

Images