Hospital specialty decision support system, method, corresponding equipment and storage medium
By constructing a causal network and linear regression model, the problem of lack of quantitative analysis in hospital specialty management was solved, and scientific management decision support was achieved.
Patent Information
- Application Number
- CN202111044134.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-07
- Publication Date
- 2025-10-10
- Estimated Expiration
- 2041-09-07
AI Technical Summary
In existing technologies, hospital specialty operations rely on personal management experience, lack quantitative analysis and historical data mining, and are unable to make forward-looking operational decisions.
By establishing a causal network, calculating the correlation coefficients between indicators and original variables, building a node hierarchical tree structure, establishing a linear regression statistical model, and optimizing the model, business decision support is automatically output.
It provides a quantitative business decision support system to help hospitals understand the business status of their specialties, support vertical and horizontal comparisons, and improve the scientificity and timeliness of decision-making.
Smart Images
Figure CN113947278B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of electronic digital data processing, and in particular to a hospital specialty decision support system. The present application also relates to a hospital specialty decision support method and corresponding computer equipment and computer-readable storage medium. Background Art
[0002] As the fundamental operating unit of a hospital, clinical specialties significantly influence the acquisition of a hospital's competitive advantage. Specialty management has become a crucial path for advancing hospital management. From an economic and management perspective, specialty management requires the integrated utilization of exogenous factors (such as endowments, positioning, and structure) and endogenous factors (such as resources, capabilities, and knowledge), integrating internal and external factors to build a hospital governance system tailored to the development of specialties. This approach maximizes the utilization of specific operational functions such as performance management, equipment management, material supply, and human resources. Therefore, identifying the core factors and relevant indicators that drive the strategic development of specialty management plays a decisive role in fostering a hospital's competitive advantage. However, as medical institutions evolve towards scale, specialization, and complexity, many specialty management practices still rely on individual management experience, employing qualitative rather than quantitative analysis. This approach fails to leverage historical operational data to gain valuable insights and inform forward-looking operational decisions. Summary of the Invention
[0003] In order to overcome the deficiencies in the prior art, the present invention provides a hospital specialty decision support system, method, corresponding equipment and storage medium, which can effectively conduct hospital data mining and automatically output mining results, providing decision support for hospital specialty management.
[0004] In a first aspect of the present invention, a hospital specialty decision support system is provided, comprising:
[0005] A correlation calculation module is used to calculate the pairwise causal correlation coefficients between indicators, between original variables, and between indicators and original variables based on the indicators of the pre-established indicator set and the original variables of the hospital data system, and to determine the causal relationship network between the indicators and the original variables;
[0006] an indicator and original variable allocation module, configured to allocate the indicators and original variables to corresponding nodes of the node hierarchical tree structure according to the pairwise causal correlation coefficients between the indicators, the original variables and the node indicator representatives in the node hierarchical tree structure pre-established according to the specialized subject, and the causal relationship network, wherein each node in the node hierarchical tree structure has been pre-assigned an indicator selected from the indicator set as an indicator representative;
[0007] A model building module is used to build a linear regression statistical model for each node based on the node hierarchical tree structure of the assigned indicators and original variables;
[0008] a model optimization module, configured to optimize the linear regression statistical models respectively, remove model variables with relatively small influence, and obtain final statistical models;
[0009] a decision question analysis module, configured to receive and analyze a specialty decision question;
[0010] a decision result output module, configured to automatically output a result of the specialty decision question based on the analyzed specialty decision question, the index data, the pairwise causal correlation coefficients, the node hierarchical tree structure, and / or the final statistical models.
[0011] In a second aspect of the present application, a hospital specialty decision support method is provided, comprising:
[0012] calculating pairwise causal correlation coefficients between indexes, between original variables, and between indexes and original variables based on indexes of a pre-established index set and original variables of a hospital data system, and determining a causal relationship network of the indexes and the original variables;
[0013] allocating the indexes and the original variables to corresponding nodes of a node hierarchical tree structure based on the pairwise causal correlation coefficients and the causal relationship network, wherein each node of the node hierarchical tree structure has been pre-allocated an index selected from the index set as a node index representative;
[0014] establishing a linear regression statistical model for each node based on the node hierarchical tree structure of the allocated indexes and original variables;
[0015] optimizing the linear regression statistical models respectively, removing model variables with relatively small influence, and obtaining final statistical models;
[0016] receiving and analyzing a specialty decision question;
[0017] automatically outputting a result of the specialty decision question based on the analyzed specialty decision question, the index data, the pairwise causal correlation coefficients, the node hierarchical tree structure, and / or the final statistical models.
[0018] In a third aspect of the present application, a computer device is provided, comprising a processor, a memory, and a computer program stored on the memory and executable on the processor, wherein the processor implements the functions of the system according to the first aspect of the present application or implements the steps of the method according to the second aspect of the present application when executing the computer program.
[0019] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the functions of the system according to the first aspect of the present invention or implements the steps of the method according to the second aspect of the present invention.
[0020] According to the present invention, by determining the causal relationship network of indicators and original variables and calculating the pairwise causal correlation coefficients between indicators and original variables, and on this basis allocating the indicators and original variables to the corresponding nodes of the hierarchical tree structure of the specialty management theme nodes, the indicator hierarchical regression modeling method is used to systematically analyze the logical relationship between the hospital management indicators and the original variables, explore the influencing factors that are of great significance to the effectiveness of the management strategy, provide relevant personnel with a clear and simple indicator network map, and form a comprehensive, adaptable and inherently related management decision-making system by analyzing specialty decision-making questions and automatically outputting results. The system is suitable for both vertical comparison of specialty management (different periods in the same department) and horizontal comparison of specialty management (different departments in the same hospital in the same period, and the same department in different hospitals in the same period), which helps relevant personnel to understand the specialty management status in a timely manner through comparison and make effective decisions.
[0021] Other features and advantages of the present invention will become more apparent after reading the detailed description of the embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 is a block diagram of an embodiment of a system according to the present invention;
[0023] Figure 2 An example of a hierarchical tree structure of nodes for a topic;
[0024] Figure 3 for Figure 2 A specific application example of the node hierarchical tree structure shown;
[0025] Figure 4 is an example of a causal network between indicators;
[0026] Figure 5 An example of a causal relationship network between specific indicators and original variables is shown;
[0027] Figure 6 Example of a hierarchical tree structure for nodes that have been assigned indicators and original variables;
[0028] Figure 7 for Figure 6 Specific application examples of the tree structure shown;
[0029] Figure 8 FIG. 1 is a flow chart of an embodiment of a method according to the present invention.
[0030] For clarity, these drawings are diagrammatic and simplified for clarity, and only show details that are essential for an understanding of the present application, with other details being omitted. DETAILED DESCRIPTION
[0031] Embodiments and examples of the present application will now be described in detail with reference to the accompanying drawings.
[0032] The scope of the present application will be apparent from the detailed description given below. However, it is to be understood that the detailed description and specific examples, while indicating preferred embodiments of the application, are given by way of illustration only.
[0033] Figure 1 A block diagram of a preferred embodiment of a hospital specialty decision support system according to the present application is shown. The specialty decision support system includes a correlation calculation module 102, an index and raw variable assignment module 104, a model building module 106, a model optimization module 108, a decision question resolution module 110, and a decision result output module 112.
[0034] The correlation calculation module 102 is used to calculate the pairwise causal correlation coefficients between the indices of a pre-established index set and the raw variables of a hospital data system, and to determine the causal relationship network of the indices and raw variables, based on the indices and raw variables. The raw variables are the data variables collected and stored in the hospital system, and the indices are the variables calculated from the raw variables, which are of interest or to be evaluated by people.
[0035] The hospital usually has a specialty management information system, and such a message system can be accessed to collect data through a database connection, and to integrate the data into the hospital data system used herein.
[0036] The collected data can include, for example, the following types:
[0037] 1) Charge item dictionary, including item code, item name, and charge category;
[0038] 2) Medical service item dictionary, including item code, item name, and item belonging to large category name;
[0039] 3) Department dictionary, including year, month, department code, and department name;
[0040] 4) Department workload, including year, month, department code, and department name;
[0041] 5) Charge item details, including time, single department code, and item code;
[0042] 6) Patient information, including patient main index, name, and birth date;
[0043] 7) Department operation information, including department service types, service volume, service charges, and number of patients; and
[0044] 8) Department expenditure information, including materials, drug costs, operating expenses, etc.
[0045] The multi-dimensional data obtained from different hospitals and departments can also be sorted and standardized, such as unifying units, formats, and removing empty data.
[0046] Based on the established hospital data system, a set of commonly used indicators can be established according to statistical purposes. For example, an indicator set may include {department income, financial subsidies, medical income, scientific and educational income, other income, inpatient income, outpatient income, surgical income, number of surgeries, average surgical fee per surgery, inpatient essential drug utilization rate, bed utilization rate, physicians / actual number of open beds, inpatient examination income, number of inpatients, outpatient number, average outpatient fee per patient, outpatient material income, average outpatient appointment rate, and outpatient examination income}.
[0047] Some indicators in the indicator set may not have direct data in the hospital data system. The data corresponding to these indicators can be calculated and saved based on the data in the hospital data system, the relationship between the indicators and the corresponding data, etc. for subsequent use.
[0048] According to each specialty theme, such as specialty business theme, for example, general surgery income theme, a logical progressive hierarchical decomposition is performed from top to bottom to form a node hierarchical tree structure, such as Figure 2 As shown in , each node represents a logical branch of a topic / node on the previous level. For each node, an indicator representing the node is assigned, called an indicator representative, and these indicator representatives are removed from the indicator set. Nodes at the same level have the characteristics of relatively independent logic and relatively consistent information. "Relatively consistent information" means that the information contained in the nodes at the same level is in the same logical structure and the information integration is consistent. For example, hospital income includes medical income, financial subsidies, scientific research income and other income, and these four incomes are at the same level; medical income includes inpatient income, outpatient and emergency income, etc., and these two incomes are at the same level (information level, consistent integration). Nodes on adjacent upper and lower levels have the characteristics of the upper level being the result and the lower level being the cause. Figure 3 The following shows an example of a hierarchical tree structure of nodes for a general surgery income topic. For simplicity, only three layers are shown.
[0049] In one embodiment, statistical correlation analysis can be used to calculate pairwise causal correlation coefficients between all indicators and original variables. Based on the causal correlation coefficients, the causal correlations between the indicators and original variables are represented as a relationship network. In one embodiment, an indicator or original variable can be selected and the indicators and original variables whose causal correlation coefficients are greater than a predetermined threshold are connected to form a relationship network. After completing one network, the above process is repeated for the remaining indicators and original variables that have not yet formed a network, thereby forming multiple relationship networks. Figure 4 An example of a causal network among indicators is shown. Figure 5 An example of a causal relationship network between specific indicator examples and original variable examples is shown.
[0050] The indicator and original variable allocation module 104 is used to allocate indicators and original variables to corresponding nodes according to the pairwise causal correlation coefficients between the indicators, original variables and node indicator representatives and the causal relationship network between the indicators and original variables.
[0051] In the node hierarchical tree structure, from bottom to top, according to the network relationship obtained by the correlation calculation module 102, the indicators and original variables that have causal correlation with the indicator representatives of the corresponding nodes are allocated to the corresponding nodes. Figure 6 As shown, starting from the bottom level of the node hierarchical tree structure, indicators, original variables, and the relationship networks they contain are assigned to corresponding nodes based on the pairwise causal correlations between the indicators, original variables, and the indicators of the corresponding nodes. For different nodes at the same level, as long as the causal correlation coefficients between the indicators, original variables, and the indicators of the corresponding nodes are above the critical value, these indicators and original variables can appear in different nodes at the same time.
[0052] For each node in the previous layer, assign indicators and original variables that are causally related to the node's indicator and that do not appear in the lower layers. In other words, after assigning indicators and original variables to the lowest node, assign indicators and original variables to the adjacent nodes in the previous layer that do not appear in the indicators assigned to the lowest node and that have a correlation coefficient with the indicator representative of the node in the previous layer greater than a critical value. This process repeats until all nodes in each layer have completed the assignment of indicators and original variables.
[0053] Figure 7 Shows the Figure 3Indicators and original variables are assigned to the lowest nodes (inpatient income and outpatient income) in the hierarchical tree structure of the general surgery revenue theme. Indicators and original variables are assigned to corresponding nodes only if there is an indicator in the network where the indicator or original variable resides, or if the causal correlation coefficient between the original variable and the indicator at that node is greater than a critical value A (for example, A = 0.25). For upper nodes in the hierarchical tree structure, indicators and original variables whose causal correlation coefficient with the indicator at that node is greater than the critical value and does not appear in lower nodes are assigned.
[0054] The model building module 106 is used to build a linear regression statistical model for each node based on the node hierarchical tree structure of the assigned indicators and original variables.
[0055] For the indicator representatives of each two adjacent layers of the node hierarchical tree structure of the assigned indicators and original variables, regression modeling is used to depict the relationship between the upper and lower layer indicator representatives. That is, from the top to the bottom of the node hierarchical tree structure, linear regression fitting is performed on the indicator representatives of each two adjacent layers of nodes and the indicators assigned to the upper layer and the original variables. Figure 6 For the example shown, we can get the expression (i.e. linear regression statistical model):
[0056] A1.1=a1*A2.1+a2*A2.2+a3*X1+a4*X2
[0057] Based on a significance analysis of whether coefficients a1, a2, a3, and a4 are equal to zero, the linear regression equation is optimized to remove non-significant original variables or indicators. Coefficients a1, a2, a3, and a4 are determined based on the type of regression model. For linear regression models, the coefficients are determined using the least squares method to minimize the difference between the fitted values and the observed values. For models such as logistic regression, the coefficients are determined using the maximum probability method to maximize the probability of the observed values being observed.
[0058] After completing the regression fitting of the nodes at the adjacent levels, for each node at the bottom layer, the node indicator representative is used as the dependent variable, and the indicator assigned to the node and the original variable are used as independent variables. Through regression analysis, a linear regression statistical model is preliminarily established. Figure 6 In the example shown, each node in the bottom layer is fitted with the indicators and original variables assigned to the corresponding nodes. The indicators represent the dependent variable Y, and the indicators and original variables are the independent variables X. The following relationship expression can be established:
[0059] A2.1=b1*A+b2*B+b3*C+b4*D+b5*E
[0060] A2.2=b6*A+b7*D+b8*F
[0061] Wherein, b1-b8 are coefficients, which are determined in a similar way as coefficients a1-a4. A, B, C, D, E, F represent indexes and original variables.
[0062] The model optimization module 108 is configured to optimize the established linear regression statistical model respectively, remove the model variables with relatively small influence, and obtain a final statistical model.
[0063] According to the linear regression statistical model quality parameters (parameters reflecting the fitting quality of the regression model, used for model debugging and screening), variable screening, variable form change, high-influence data discovery and adjustment are performed to establish a medium-term statistical model.
[0064] Specifically, through quality analysis of the fitting, variable screening, variable form conversion are performed according to R^2, AIC, p-value and other model quality parameters, and high-influence data points are found by calculating Cook's distance of each data and other methods, and sensitivity analysis such as data reduction is performed, so as to optimize the preliminary linear regression statistical model established in the foregoing, so as to improve or maintain the fitting quality of the model and make the model relatively simple, that is, the variable form on the right side of the model expression is simple and the number of variables is small, and the following medium-term statistical model (in the example of Figure 6 ) can be obtained.
[0065] A2.1=c1*A+c2*B+c3*D+c4*E
[0066] A2.2=c5*A+c6*D+c7*F
[0067] Similarly, c1-c7 are coefficients, which are determined in a similar way as coefficients a1-a4. A, B, D, E, F represent indexes and original variables.
[0068] According to the selected variables and variable forms of the medium-term statistical model obtained by optimization, principal component analysis is further applied, linear combinations of several groups of independent variables X with relatively high causal correlation with A2.1 and A2.2 are retained, so as to further reduce the bilinear influence and screen variables (that is, delete some variable combinations with small influence on the dependent variable), reduce the dimension of the equation, make the result more accurate and stable, and obtain a final statistical model. Specifically, for the index and original variable set X, singular value decomposition of the matrix is performed to obtain X=UΣV T , where U and V are matrices composed of characteristic vectors of X T and X T , and Σ is a diagonal matrix with singular points falling on the diagonal. Through linear transformation of the original index set, Z=XV T , a model with A2.1 and A2.2 as dependent variables and Z as independent variables is established. By deleting the directions with small singular values, a final statistical model (in the example of Figure 6 (in the example):
[0069] A2.1=d1*A+d2*D+d3*E
[0070] A2.2=d4*A+d5*D
[0071] Similarly, d1-d5 are coefficients, which are determined in a similar way to coefficients a1-a4. A, D, and E represent indicators and original variables.
[0072] Based on the final statistical model, relevant indicators, original variables, and coefficients can be extracted from the model. These indicators and original variables are the fundamental factors affecting the corresponding nodes, and the corresponding coefficients determine the influence (strength) of the corresponding indicators and original variables on the nodes.
[0073] In other embodiments, principal component analysis may not be applied when optimizing the established linear regression statistical models, and optimization may be performed only based on the linear regression statistical model quality parameters.
[0074] For example, for Figure 3 and Figure 7 In the example shown, the relationship between level 1 and level 2, and between level 2 and level 3 is established. Through fitting, the following two relationship equations are obtained:
[0075] Department income = 1*financial subsidy + 1*medical income + 1*scientific and educational income + 1*other income
[0076] Medical income = 1*hospitalization income + 1*outpatient income
[0077] The coefficients of these indicators and original variables (set to 1 in this example) are tested for non-zero hypothesis to determine whether to retain these indicators and original variables in the relationship equation, thereby simplifying the model. After completing the fitting of the logical relationship between the nodes of the hierarchy, statistical modeling is performed on the indicator representative of each node in the bottom layer (dependent variable) and the indicators and original variables assigned to the node (independent variables).
[0078] exist Figure 7 In the example shown, the following two mathematical relationships are obtained through preliminary data fitting:
[0079] Hospitalization income = a0 + a1 * surgical income + a2 * average surgical fee + a3 * number of surgeries + a4 * number of inpatients + a5 * hospitalization examination income + a6 * inpatient basic drug utilization rate + a7 * physicians / actual number of open beds + a8 * bed utilization rate
[0080] Outpatient income = b0 + b1 * number of outpatients + b2 * average outpatient cost + b3 * outpatient examination income + b4 * outpatient material income + b5 * average outpatient appointment rate
[0081] By analyzing the quality of the fitting, we can screen variables and transform the variable form. For example, for hospitalization income, the updated statistical model is:
[0082] Hospitalization income = c0 + c1 * average cost per surgery + c2 * number of surgeries + c3 * hospitalization examination income + c4 * basic drug utilization rate of hospitalized patients
[0083] By further processing the data matrix:
[0084]
[0085] Perform singular value decomposition (SVD), X = UΣV T , define a new index set Z = XV T Using principal component analysis, we remove transformed indicators with small singular values, and through hypothesis testing of the non-zero coefficients of indicators, we retain the indicators and original variables with significant results, thus obtaining a further model:
[0086] Hospitalization income = d0 + d1 * number of surgeries + d2 * average cost per surgery + d3 * hospitalization examination income
[0087] a0, b0, c0, and d0 are the intercepts of the straight lines, representing the value of Y when the other variable X is 0, which has little practical significance.
[0088] In the above example, we can see that the core factors influencing department revenue come from fiscal revenue, medical revenue, scientific and educational revenue, and other revenue. The primary factors influencing medical revenue come from inpatient and outpatient revenue. The fundamental factors influencing inpatient revenue, and thus department revenue, are the number of surgeries, average cost per surgery, and inpatient examination revenue. For every corresponding unit increase, inpatient revenue increases by d1, d2, or d3 units, and thus department revenue increases by d1, d2, or d3 units.
[0089] The decision question parsing module 110 is used to receive and parse specialized decision questions.
[0090] Hospital personnel can conduct influencing factor mining and ask questions related to specialized decision-making for a specific indicator or category of indicators. For example, specialized (operational) decision-making questions may include the following types: factual statistics, factor mining, and quantitative recommendations. Specialty decision-making questions can be parsed using technologies such as natural language processing to identify the specialized decision-making questions raised by hospital personnel. In other embodiments, specialized decision-making questions raised by hospital personnel can also be identified through templates or question element selection.
[0091] The decision result output module 112 is used to automatically output the results of the proposed specialty decision problem based on the indicator data, pairwise causal correlation coefficients, node hierarchical tree structure, and / or the final statistical model. In one embodiment, the results can be output according to a pre-established template corresponding to the specialty decision question.
[0092] Some non-limiting examples of various types of questions and their output are given below.
[0093] Factual statistical questions can take the following forms:
[0094] Question A1: What is [time][organization][indicator]?
[0095] Example: How many orthopedic surgeries were performed in March 2019?
[0096] Output example: 297
[0097] Question A2: What is the ranking of [time][organization][indicator] in [parent organization] (visualization option)
[0098] Example: What is the ranking of the Department of Surgery in terms of the number of orthopedic surgeries performed in March 2019?
[0099] Output result example: 4th (Orthopedics): 297 (1st (General Surgery): 403; 2nd (Cardiac Surgery): 356; 3rd (Neurology): 310)
[0100] Question A3: What is the rate of change of [organization] [indicator] from [time 1] to [time 2]?
[0101] Example: What was the rate of change in the number of orthopedic surgeries from March 2019 to May 2019 (visualization option)
[0102] Output example: 297->313 (+5.39%)
[0103] Question A4: What is the trend of [organization] [indicator] in [time period] (visualization option)?
[0104] Example: What was the trend in the number of orthopedic surgeries from January 2019 to June 2019?
[0105] Output example: 201->167(-16.92%)->297(+77.84%)->290(-2.36%)->313(+7.93%)->358(+14.38%)
[0106] Question A5: How does [time][organization][metric] compare to [organization 2,…, organization N] (N ≥ 2)? (Visualization option)
[0107] Example: What is the output result of comparing the number of surgeries in Department of Orthopedics I with those in Department of Orthopedics II and Department of Orthopedics III in July 2019? Example: Department of Orthopedics I: 188 (Department of Orthopedics II: 150; Department of Orthopedics III: 83)
[0108] The "visualization options" above mean that the corresponding output results can be displayed in the form of curve graphs, bar charts, etc.
[0109] For factual statistical questions, the system queries, extracts, and calculates database data, outputting numerical results. If no results are found, the system outputs "No results." For example, for question A4, the database is queried for the department "Orthopedics" and the number of surgeries from January to June 2019. Based on the number of surgeries found, the system then calculates the trend. For example, the trend from January to February is the percentage of (number of surgeries in February / number of surgeries in January - 1).
[0110] Factor mining problems can include the following forms, which are used to conduct factor mining for the above-mentioned fact statistics:
[0111] Question B1: The key influencing factors (indicators) are:
[0112] Question B 2: The key factors (indicators) that influence its current ranking are:
[0113] Question B 3: The key factors (indicators) that lead to its current rate of change are:
[0114] Question B 4: The key factors (indicators) that have led to the current trend of change are:
[0115] Question B 5: The key factors (indicators) that lead to the target comparison results are:
[0116] Output result example (question B1): Number of hospitalizations (355) <- Bed occupancy rate (91.28%) <- Average length of stay (8.3) <- Department CMI value (1.02)
[0117] The output examples for questions B2 through B5 are similar to the examples above and also include comparative data for other time periods or organizations.
[0118] As mentioned above, all models are fitted using the indicator represented by Y. For factor mining problems, factor mining can be performed using the final statistical model and corresponding data of the corresponding node. For example, if the relevant indicators are located at level l (1≤l≤L) of a hierarchical tree structure with L nodes, the key influencing factors are decomposed and ranked according to the corresponding coefficients in the [l:L] layer. (L-l+1) indicators and their numerical results are then output in sequence. For time dimension comparison problems, the numerical results for other time dimensions are also output. For organizational dimension comparison problems, (L-l+1) indicators and their numerical results for each organizational dimension are output in sequence based on the comparison of network-level indicators and coefficients. In other words, if an indicator is at the upper level of the tree hierarchy, all lower-level indicators and factors are displayed layer by layer. For example, if the medical revenue of Orthopedics Department 1 decreased in January compared to February, the results of the next level below are output first, such as the decrease in medical revenue was mainly due to the decrease in inpatient revenue. The next level below this is output, such as the decrease in inpatient revenue was due to the decrease in the number of surgeries. Finally, the results of the final level are output, such as the decrease in the number of surgeries was due to a decrease in hospitalizations and an increase in the average length of stay.
[0119] For questions B1, B2, B3, B4, and B5 above, let's use the following example as an example. Assume that the final statistical model is:
[0120] Number of surgeries = 0.6*number of hospitalizations + 12*bed occupancy rate - 3.6*average length of stay + 16*department CMI
[0121] The data for the Department of Orthopedics in January were: 200 surgeries, 300 hospital admissions, 95% bed occupancy, 4 average length of stay, and a CMI of 1.44;
[0122] The data for the Department of Orthopedics in February were: 150 surgeries, 250 hospital admissions, 80% bed occupancy, 10 average length of stay, and a CMI of 1.65;
[0123] The data for the Orthopedics Department II in January were: 220 surgeries, 310 hospital admissions, 90% bed occupancy, 3.5 average length of stay, and a CMI of 2.24;
[0124] The data for Department of Orthopedics II in February were: 190 surgeries, 290 hospitalizations, 83% bed occupancy rate, 4 average length of stay, and a CMI of 1.28.
[0125] Question B1: From the final statistical model, it can be concluded that the key factors affecting the number of surgeries are: number of hospitalizations, bed occupancy rate, average length of stay, and department CMI.
[0126] Question B2: The key factors that led to the number of surgeries in Department of Orthopedics I being lower than that in Department of Orthopedics II in January were: CMI (difference: (2.24-1.44)*coefficient 16=12.8), number of hospitalizations (difference 10*coefficient 0.6=6), average length of stay (difference 0.5*coefficient 3.6=1.8) and bed occupancy rate (difference 0.05*coefficient 12=0.6).
[0127] Question B3: The key factors that led to the decrease in the number of surgeries in the Department of Orthopedics in February compared with January were: number of hospitalizations (difference 50*coefficient 0.6=30), average length of stay (difference 6*coefficient 3.6=21.6), CMI (difference 0.21*coefficient 16=3.36) and bed occupancy rate (difference 0.15*coefficient 12=1.8).
[0128] Asking questions of type B4 and B5 is similar to asking questions of type B2 and B3, and also involves comparisons between two items.
[0129] Quantitative suggestion questions may include the following:
[0130] Question C1: The goal is to achieve a [percentage] [change (increase / decrease)] of [organization] [metric] in [time period]. What suggestions (visualization options) can be provided:
[0131] Example: If the goal is to increase the number of orthopedic surgeries by 10% in October 2021, the following recommendations might be provided:
[0132] Output: Suggest changes of [percentage 1,…,percentage M] to [index 1,…,index M] (M ≥ 1) respectively
[0133] Output result example: It is recommended to make changes of +3.50%, -7.35%, +6.73%, and +12.10% to the department's CMI value, average length of stay, bed occupancy rate, and number of hospitalizations, respectively.
[0134] For quantitative suggestion-type questions, causal logical reasoning is performed through the model building module 106 and the model optimization module 108. According to the model fitting results and the corresponding coefficient rankings, the percentage change of key influencing factors is obtained and the corresponding numerical results are output.
[0135] For example, the fact statistics type, factor mining type, and quantitative suggestion type questions are as follows:
[0136] Factual statistics: What is the rate of change in a general surgery medical income from January 2021 to June 2021?
[0137] Factor mining type: The key factors (indicators) that lead to its current rate of change are:
[0138] Quantitative Recommendation: The goal is to increase the revenue of a general surgery company by 5% by October 2021. The following recommendations can be provided:
[0139] The corresponding output results are as follows:
[0140] Output result (factual statistics): 603->667 (+10.61%)
[0141] Output results (factor mining type): Number of hospitalizations (713->802(+12.48)) <- Bed occupancy rate (89.20%->90.35%(+1.29%)) <- Average length of stay (8.8->8.3(-6.02%)) <- Department CMI value (0.89->1.05(+17.98))
[0142] Output results (quantitative suggestion type): It is recommended to make changes of +2.76%, -3.83%, +2.90%, and +8.75% to the department's CMI value, average length of stay, bed occupancy rate, and number of hospitalizations, respectively.
[0143] Figure 8 A flowchart of a preferred embodiment of the hospital specialty decision support method according to the present invention is shown.
[0144] In step S802, based on the indicators of the pre-established indicator set and the original variables of the hospital data system, the pairwise causal correlation coefficients between the indicators, between the original variables, and between the indicators and the original variables are calculated to determine the causal relationship network between the indicators and the original variables;
[0145] In step S804, based on the pairwise causal correlation coefficients between the indicators, the original variables, and the node indicator representatives in the node hierarchical tree structure pre-established by subject, and the causal relationship network, the indicators and the original variables are assigned to corresponding nodes in the node hierarchical tree structure, wherein each node in the node hierarchical tree structure has been pre-assigned an indicator selected from the indicator set as an indicator representative;
[0146] In step S806, a linear regression statistical model is established for each node based on the node hierarchical tree structure of the assigned indicators and original variables;
[0147] In step S808, the linear regression statistical model is optimized respectively, and model variables with relatively small influence are removed to obtain a final statistical model;
[0148] In step S810, a specialist decision-making question is received and parsed;
[0149] At step S812, according to the parsed specialty decision question, the result of the specialty decision question is automatically outputted based on the index data, the pairwise correlation coefficient, the node hierarchical tree structure, and / or the final statistical model.
[0150] In another embodiment, the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the function of the system embodiment shown and described or other corresponding system embodiment or implements the steps of the method embodiment shown and described or other corresponding method embodiment. Figure 1-7 In another embodiment, the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the function of the system embodiment shown and described or other corresponding system embodiment or implements the steps of the method embodiment shown and described or other corresponding method embodiment. Figure 8 In another embodiment, the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the function of the system embodiment shown and described or other corresponding system embodiment or implements the steps of the method embodiment shown and described or other corresponding method embodiment.
[0151] In another embodiment, the present application provides a computer device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, wherein the processor implements the function of the system embodiment shown and described or other corresponding system embodiment or implements the steps of the method embodiment shown and described or other corresponding method embodiment when executing the computer program. Figure 1-7 In another embodiment, the present application provides a computer device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, wherein the processor implements the function of the system embodiment shown and described or other corresponding system embodiment or implements the steps of the method embodiment shown and described or other corresponding method embodiment when executing the computer program. Figure 8 In another embodiment, the present application provides a computer device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, wherein the processor implements the function of the system embodiment shown and described or other corresponding system embodiment or implements the steps of the method embodiment shown and described or other corresponding method embodiment when executing the computer program.
[0152] The various embodiments described herein, or certain portions thereof, can be combined, and the order of steps can be modified, as appropriate, without departing from the scope of the application. Further, the various aspects of the application can be implemented using software, hardware, firmware, or a combination thereof and / or other computer- implemented modules or devices performing the described functions. Software implementations of the application could comprise executable code stored on a computer-readable medium which is executable by one or more processors. The computer-readable medium could comprise a computer hard disk drive, ROM, RAM, flash memory, portable computer storage media such as a CD-ROM, a DVD-ROM, a flash drive, and / or other storage devices and / or computer memory with a universal serial bus (USB) interface, and / or any other appropriate tangible or non-transitory computer-readable medium or computer memory on which executable code can be stored and executed by a processor. The application can be used in conjunction with any appropriate operating systems.
[0153] Unless otherwise specified, the singular forms "a", "an", and "the" used herein include the plural meaning (i.e., having "at least one"). It should be further understood that the terms "having", "including", and / or "comprising" used in this specification indicate the presence of the recited features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. As used herein, the term "and / or" includes any and all combinations of one or more of the listed items.
[0154] While some preferred embodiments of the present invention have been described above, it should be emphasized that the present invention is not limited to these embodiments and may be implemented in other ways within the scope of the present invention. Those skilled in the art may make various variations and modifications to the present invention based on the technical concept of the present invention and without departing from the scope of the present invention, and such variations and modifications shall still fall within the scope of protection of the present invention.
Claims
1. A hospital specialty decision support system, characterized by: The system comprises: A correlation calculation module is used to calculate the pairwise causal correlation coefficients between indicators, between original variables, and between indicators and original variables based on the indicators of the pre-established indicator set and the original variables of the hospital data system, and determine the causal relationship network between the indicators and the original variables according to the calculated causal correlation coefficients; an indicator and original variable allocation module, configured to allocate the indicators and original variables to corresponding nodes of the node hierarchical tree structure according to the pairwise causal correlation coefficients between the indicators, the original variables and the node indicator representatives in the node hierarchical tree structure pre-established according to the specialized subject, and the causal relationship network, wherein each node in the node hierarchical tree structure has been pre-assigned an indicator selected from the indicator set as an indicator representative; The model building module is used to perform linear regression fitting on the indicator representatives of the nodes in each two adjacent layers and the indicators assigned to the upper layer and the original variables based on the node hierarchical tree structure with assigned indicators and original variables from top to bottom. After completing the linear regression fitting of the nodes in the adjacent layers, for each node in the bottom layer, the indicator representative of the node is used as the dependent variable, and the indicator assigned to the node and the original variable are used as independent variables. Through regression analysis, a linear regression statistical model is established; A model optimization module is used to optimize the linear regression statistical model respectively, remove model variables with relatively small influence, and obtain a final statistical model; Decision question parsing module, used to receive and parse specialist decision questions; The decision result output module is used to automatically output the result of the specialized decision problem according to the analyzed specialized decision problem based on the indicator data, the pairwise causal correlation coefficient, the node hierarchical tree structure, and / or the final statistical model.
2. The system according to claim 1, wherein: The indicator and original variable allocation module includes: The removal submodule is used to remove the indicator representative from the indicator set; The allocation submodule is used to allocate the indicators and original variables to the corresponding nodes from bottom to top for each layer of the node hierarchical tree structure, according to the remaining indicators in the indicator set after the indicator representatives have been removed, the original variables and the indicator representatives of each node in each layer, wherein, for the nodes of the upper layer, the indicators and original variables that have a causal correlation with the node but do not appear in the lower layer are allocated.
3. The system according to claim 2, characterized in that When assigning indicators and original variables, only those indicators and original variables whose causal correlation coefficients with the indicator representatives of the corresponding nodes are higher than a predetermined critical value are assigned.
4. The system according to claim 1, wherein: The system further comprises: The indicator data calculation module is used to calculate the data of the indicators in the indicator set that have no corresponding data in the hospital data system.
5. The system according to claim 1, wherein: The decision result output module outputs the result according to a pre-established template corresponding to the specialty decision question.
6. The system according to claim 1, wherein: The decision result output module includes: when the analyzed specialty decision problem is of factor mining type, decomposing the key influencing factors from the final statistical model represented by the indicators involved, sorting the key influencing factors according to the coefficients corresponding to the key influencing factors, and outputting the key influencing factors and corresponding data in order according to the sorting results.
7. The system according to claim 6, characterized in that The decision result output module also includes: calculating the difference value according to the dimensions involved based on the data and coefficients of each key influencing factor, sorting the key influencing factors according to the calculated difference values, and outputting the key influencing factors and difference values of the corresponding dimensions in order according to the sorting results.
8. A hospital specialty decision support method, characterized by: The method comprises: Based on the indicators of the pre-established indicator set and the original variables of the hospital data system, the pairwise causal correlation coefficients between the indicators, between the original variables, and between the indicators and the original variables are calculated, and the causal relationship network between the indicators and the original variables is determined according to the calculated causal correlation coefficients; According to the pairwise causal correlation coefficients between the indicators, the original variables and the node indicator representatives in the node hierarchical tree structure pre-established by subject and the causal relationship network, the indicators and the original variables are assigned to corresponding nodes in the node hierarchical tree structure, wherein each node in the node hierarchical tree structure has been pre-assigned an indicator selected from the indicator set as an indicator representative; Based on the node hierarchical tree structure of the assigned indicators and original variables, linear regression fitting is performed on the indicator representatives of the nodes in each two adjacent layers and the indicators assigned to the upper layer and the original variables from top to bottom of the node hierarchical tree structure. After completing the linear regression fitting of the nodes in the adjacent layers, for each node in the bottom layer, the indicator representative of the node is used as the dependent variable, and the indicator assigned to the node and the original variable are used as the independent variables. Through regression analysis, a linear regression statistical model is established; Optimizing the linear regression statistical model respectively, removing model variables with relatively small influence, and obtaining a final statistical model; Receive and resolve specialist decision-making inquiries; According to the analyzed specialty decision problem, the result of the specialty decision problem is automatically output based on the indicator data, the pairwise causal correlation coefficients, the node hierarchical tree structure, and / or the final statistical model.
9. A computer device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to claim 8 when executing the computer program.
10. A computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the method according to claim 8 when executed by a processor.