A clinical pathway variation detection method based on Petri net synchronous synthesis operation

Abstract

The present invention discloses a clinical pathway variation detection method based on Petri net synchronous synthesis operation, which relates to the technical field of data detection and calculation. It solves the technical problem of low efficiency in detecting clinical pathway variations by using business alignment means in the prior art. The detection method includes the following steps: First, screen the traces in the clinical pathway diagnosis and treatment log, remove the noise, and retain the valid traces; then, based on the valid traces and their corresponding process models, establish a new synthesis model, and the new synthesis model can reflect the deviation between the diagnosis and treatment log and the clinical pathway; finally, use the A<supgt;*< / supgt; algorithm with a time complexity positively correlated with the complexity of the selected synthesis model to find the optimal alignment. The method of the present invention can greatly reduce the time consumed by alignment. The experimental data in the embodiments of the present invention show that this algorithm has a significantly improved efficiency compared with other variation detection algorithms, proving the superiority and effectiveness of this algorithm in variation detection.

Description

Technical Field

[0001] The present invention relates to the field of data detection and computing technology, and in particular to a clinical pathway variation detection method based on Petri net synchronous synthesis operation. Background Art

[0002] In recent years, with the intensification of the aging population, the medical system is facing increasing pressure. Therefore, the adoption of clinical pathways that can standardize medical behavior, improve medical quality, and control medical costs has become an inevitable choice for my country's medical reform. With the strong support of the state, clinical pathways have now entered the stage of large-scale promotion.

[0003] The essence of clinical pathways is to adopt standardized processes for the diagnosis and treatment of a certain disease. However, in the actual implementation of clinical pathways, there may be problems such as patients not understanding the implementation standards of clinical pathways and medical staff not being very enthusiastic about implementing clinical pathways. The above problems have greatly affected the promotion of clinical pathways. In addition, due to the imperfection of clinical pathways and sudden changes in patients' conditions, unexpected behaviors may occur in the diagnosis and treatment process of clinical pathways. Such behaviors are collectively referred to as variations. Under pressure from many aspects, the optimization and improvement of clinical pathways are imminent. The purpose of process mining is to use the effective information of event logs to discover and optimize business processes, and to use process mining methods to detect variant behaviors, which is conducive to predicting the direction of patient diagnosis and treatment processes, discovering defects in clinical pathways, and providing a basis for judgment on intervention behaviors in the diagnosis and treatment process, thereby improving clinical pathways.

[0004] With the widespread application of medical information systems, a large number of logs generated during the diagnosis and treatment process are digitized, and electronic medical records are also widely used in hospitals. The means of mining variant behaviors has also changed from a prospective method that requires manual recording to a retrospective method that is automatically processed by computers.

[0005] At present, there are many formal methods for describing process models from the perspective of control flow, such as BPMN (Business Process Modeling Notation), C-net (Casual net), EPC (Event-driven Process Chain) and Petri net, etc. Among them, Petri net is the most mature and most deeply studied process modeling language, which can concisely and intuitively describe and analyze concurrent, asynchronous, distributed, parallel and systems containing uncertain information.

[0006] Business alignment is one of the most advanced compliance checking methods. Using business alignment can effectively detect variations in clinical pathways. Compliance checking of complex models has always been a challenging research topic. Adriansyah et al. proposed a method to combine the process model and the log model into a product model and find the optimal alignment based on this. Cook et al. proposed a method to compare the trace and the process model by quantifying the similarity to obtain the alignment result. Song et al. proposed to use heuristic methods and trace replay technology to achieve alignment between the trace and the process model. This method reduces the complexity of the search space.

[0007] Although the above-mentioned prior art has made certain progress in the research of business alignment, there is still room for improvement in the efficiency of business alignment. Summary of the invention

[0008] The purpose of the present invention is to provide a clinical pathway variation detection method based on Petri net synchronous synthesis operation, which establishes a process model of the clinical pathway and uses business alignment to perform variation detection on the business process execution record and the process model. The detection method can effectively reduce the complexity of the model to improve the alignment efficiency.

[0009] In order to achieve the above objectives, the main technical difficulties that the present invention needs to overcome are how to find the optimal alignment based on the screened diagnosis and treatment logs, discover possible problems in the traces, and how to determine the efficiency index for calculating the optimal alignment.

[0010] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0011] A clinical pathway variation detection method based on Petri net synchronous synthesis operation includes the following steps in sequence:

[0012] S1. Screen the traces in the clinical pathway diagnosis and treatment log, remove noise, and retain valid traces;

[0013] S2. Establishing a new synthesis model based on the effective trace and its corresponding process model, wherein the new synthesis model reflects the deviation between the diagnosis and treatment log and the clinical pathway;

[0014] S3, using A whose time complexity is positively correlated with the complexity of the selected synthesis model * The algorithm finds the optimal alignment.

[0015] The beneficial technical effects directly brought about by the above technical solution are:

[0016] The present invention first pre-processes the diagnosis and treatment log to save time for subsequent work. Then, considering that the complexity of the reachable graph of the Petri net will increase with the increase of the scale of the Petri net, and alignment is actually based on the reachable graph to find the shortest path in logic, reducing unnecessary changes in the synthetic model, even if the number of changes is only slightly reduced, the alignment efficiency can still be greatly improved. Therefore, the present invention adopts a synchronous synthetic model for alignment calculation, which greatly reduces the time consumed by alignment. The method of the present invention can be applied to online variation detection of clinical pathways.

[0017] As a preferred embodiment of the present invention, in step S1, when screening the traces in the clinical pathway diagnosis and treatment log, all activities in the trace are checked one by one, and the sequential relationship between the currently checked activity and its subsequent activities is compared with the sequential relationship in the precedence relationship matrix. If they do not match, the inspection is terminated, the trace is designated as noise, and removed from the log.

[0018] As another preferred solution of the present invention, in step S1, a parallel activity set S is set. Before starting the traversal, the first activity in the trace is placed in S. The current activity in the traversal is set as activity cur, and the subsequent activity of activity cur in the trace is set as activity post. The rules are as follows:

[0019] S11. If there is one or more activities in the set S whose precedence relation matrix allows the occurrence time to be earlier than the activity post and vice versa, delete these activities from the set and then put the activity post into the set S;

[0020] S12. If all activities in the set form a parallel relationship with the activity post, only the activity post is placed in the set S without deleting any elements;

[0021] S13. If there is any activity in the set S that has a sequence deviation relationship with the activity post, then the trace is regarded as noise and can be directly discarded.

[0022] Further preferably, in step S2, when establishing a new synthetic model, the effective trace is first converted into a log model in the form of a Petri net, and then two transitions corresponding to the same activity are merged, and the front set and the back set are merged separately to obtain a new synthetic model.

[0023] Preferably, in step S3, the reachable graph of the new synthetic model is first calculated, and then based on A * The algorithm searches for the optimal alignment.

[0024] Preferably, the alignment result is evaluated by introducing a cost function lc((a, t)), which is used to calculate the cost of each step of movement. Among the multiple alignment results calculated, the alignment with the smallest sum of cost values ​​is the optimal alignment.

[0025] For any movement (a, t)∈γ in the alignment, the cost function is as follows:

[0026] S31, (a, t) is log movement, then lc((a, t)) = 1;

[0027] S32, (a, t) is the model movement, then lc((a, t)) = 1;

[0028] S33, (a, t) is synchronous movement, then lc((a, t)) = 0;

[0029] S34, (a, t) is an illegal move, then lc((a, t)) = +∞.

[0030] Preferably, for any movement (a, t)∈γ in the alignment, there are four possibilities:

[0031] S35. If a∈A, t=>>, (a, t) is called a log move;

[0032] S36. If a=>>, t∈T, (a, t) is called model movement;

[0033] S37. If a∈A, t∈T, (a, t) is called synchronous movement;

[0034] S38. The rest are called illegal moves.

[0035] Among them, A is an activity set.

[0036] Illegal moves are actually a special type of moves that exist in theory but will not appear in the actual business process processing. Therefore, illegal moves do not belong to the research scope of the present invention and are ignored in actual research.

[0037] Another object of the present invention is to provide an application of a clinical pathway variation detection method based on Petri net synchronous synthesis operation in a computer, wherein the application is to apply the clinical pathway variation detection method based on Petri net synchronous synthesis operation to a program running on a processor of a computer.

[0038] Another object of the present invention is to provide a computer, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the computer implements the clinical pathway variation detection method based on Petri net synchronous synthesis operation as described above.

[0039] Another object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes the clinical pathway variation detection method based on Petri net synchronous synthesis operation as described above.

[0040] Compared with the prior art, the present invention brings the following beneficial technical effects:

[0041] A proposed in the prior art * The alignment algorithm is a product synthesis algorithm that can multiply traces and process models and has a wide range of applications. Compared with the previous manual forward-looking methods, it has made a qualitative leap. However, due to the high complexity of the model, the efficiency is still unsatisfactory when calculating the optimal alignment. Therefore, the present invention proposes a clinical pathway variation detection method based on Petri net synchronous synthesis operation, which can greatly reduce the time consumed by alignment.

[0042] Finally, the present invention conducted an experimental analysis using the clinical pathway of cardiovascular diseases under the COVID-19 pandemic as an example, proving that the variation detection algorithm proposed in the present invention has the advantage of higher efficiency than traditional algorithms, and can greatly shorten the time consumed in the variation detection process. In subsequent work, this method can be applied to online variation detection. Online variation detection can promptly discover and remind patients of variations in the currently completed diagnosis and treatment process, or estimate possible subsequent variations, and provide a basis for medical personnel to conduct manual intervention to avoid adverse events. Online variation detection has higher time requirements, which can better reflect the efficiency advantages of the method proposed in the present invention.

[0043] The experimental data in the embodiments of the present invention show that the efficiency of the algorithm is greatly improved compared with other variation detection algorithms, which proves the superiority and effectiveness of the algorithm in variation detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described below in conjunction with the accompanying drawings:

[0045] Figure 1 The Petri net model of the clinical pathway for ischemic stroke pm ;

[0046] Figure 2 For model N pm The reachability graph TS pm ;

[0047] Figure 3 Synthetic diagram for change;

[0048] Figure 4 For the log model N lm ;

[0049] Figure 5 Synthesized model N sc ;

[0050] Figure 6 For model N sc The reachability graph TS sc ;

[0051] Figure 7 To provide clinical pathways for cardiovascular diseases during the COVID-19 pandemic;

[0052] Figure 8 This is a comparison chart of the number of transitions of the two models;

[0053] Fig. 9 This is a comparison chart of the arc numbers of the two models;

[0054] Fig.10 This is a comparison chart of the average number of queued nodes (reachable identifiers) of the two models;

[0055] Fig.11 This is a comparison chart of the average time consumption of the two algorithms. DETAILED DESCRIPTION

[0056] The present invention proposes a clinical pathway variation detection method based on Petri net synchronous synthesis operation. In order to make the advantages and technical solutions of the present invention clearer and more specific, the present invention is described in detail below in conjunction with specific embodiments.

[0057] 1. To facilitate understanding of the present invention, the following explanations are first given to the concepts of traces and diagnosis and treatment logs, Petri net systems, reachability graphs, and alignment mentioned in the present invention.

[0058] Definition 1: Traces and medical logs

[0059] Let A be an activity set. If there exists a process instance, that is, an activity sequence σ∈A * , then σ is called a trace. If there exists a non-empty multiset L∈β(A * ), then L is called a diagnosis and treatment log. * represents the set of all finite sequences on set A; β(A * ) represents the set A * The set of all multisets above.

[0060] A Petri net is a directed bipartite graph containing two elements, called places and transitions. The directed line between the two elements is called an arc, or flow relationship. The state of a Petri net is called a flag.

[0061] Definition 2: Petri Net System

[0062] Let A be an activity set. Define the Petri net system on set A as a tuple N = (P, T; F, α, m i ,m f ), where: P is the finite set of all libraries, T is the finite set of transitions, and The set of directed arcs that represent the relationship between places and transitions is called a flow relation; α:T→A τ is the mapping function between transition and label, where τ is the invisible transition, A τ =A∪{τ}; the identifier is the state of the Petri net, which is a multiset of the place set. i ,m f ∈β(P) are the initial and terminal identifiers of the Petri net N respectively.

[0063] like Figure 1 For a simple Petri net N pm , represents the clinical pathway of ischemic stroke. pm ={p 1 ,p 2 ,p 3 ,p 4 ,p 5}, transition set T pm ={t 1 ,t 2 ,t 3 ,t 4 ,t 5}, flow relation set F pm ={(p 1 ,t 1 ),(t 1 ,p 2 ),(p 2 ,t 2 ),(p 2 ,t 3 ),(t 2 ,p 3 ),(t 3 ,p 4 ),(p 3 ,t 4 ),(t 4 ,p 4 ),(p 4 ,t 5 ),(t 5 ,p 5 )}, the mapping function correspondence between transition and label is shown in Table 1, the initial identification m i,pm =[p 1 ], termination mark m f,pm =[p 5 ].

[0064] Table 1 Activity and transition correspondence table

[0065]

[0066] For any reachable state m∈β(P) and any transition t∈T, if and only if When it is established, t is called enabled under the label m, denoted as m[t>; at this time, a new label m′ can be obtained after the transition t occurs, and It is denoted as m[t>m′. The set of all states reachable from m is denoted as R(m), and it is agreed that m∈R(m). R(m i ) is the set of all reachable states of Petri net N.

[0067] Based on the transition triggering rules in Definition 2 and the reachable states of the Petri net, the following definition of the reachable graph of the Petri net can be given.

[0068] Definition 3: Reachable Graph

[0069] Let A be an activity set. N=(P,T;F,α,m i ,m f ) is a Petri net. TS=(S,A′,T′) is a transition system, where S=R(m i ), A′=A, TS is called the reachability graph of the Petri net N.

[0070] Figure 2 N pm The reachability graph TS pm , each node in the graph represents N pm Each arc represents a transition t that enables the predecessor m. The arrow points to the new mark m′ after the transition t occurs. The initial state of the reachable graph is m. i,pm =[p 1 ], the end state is m f,pm =[p 5 ].

[0071] When the traces in the medical log are replayed on the model, there may be a situation where the two do not fit completely. This state of misfit is called deviation, or variation, and alignment can detect deviation.

[0072] Definition 4: Alignment

[0073] Let A be an activity set. σ∈A * is the trace on A, N=(P,T;F,α,m i ,m f ) is a Petri net. The alignment between σ and A is γ∈(A>>×T>>)* is a move sequence (where >> means no move, A >>=A∪{>>}) and satisfies the following conditions:

[0074] ① The projection of the first column of the moving sequence on A (ignore >>) produces a trace, denoted by π 1 (γ)↓ A =σ;

[0075] ② The projection of the second column elements of the moving sequence onto T (ignoring >>) produces a complete triggering sequence, denoted as

[0076] For any movement (a, t)∈γ in the alignment, there are four possibilities:

[0077] ①If a∈A, t=>>, (a, t) is called a log move;

[0078] ②If a=>>, t∈T, (a,t) is called model movement;

[0079] ③If a∈A, t∈T, (a, t) is called synchronous movement;

[0080] ④The rest are called illegal moves.

[0081] According to Definition 4, alignment is a sequence of moves, and moves are the associations between activities in the trace and transitions in the model. For ① log moves, it means that an activity in the trace cannot be executed in the model; for ② model moves, it means that during the replay process, the transitions in the model are not observed in the trace; for ③ synchronous moves, it means that the activities in the trace correspond to the transitions in the model; for ④ illegal moves, illegal moves will not occur in the actual business process processing, so this type of move is ignored in this method study. Among them, ① and ② are the misfits between the trace and the process model, which reflect the deviation in alignment.

[0082] For a given trace and a process model, multiple different alignment results may be calculated. In order to obtain the most reasonable alignment result, it is necessary to introduce a cost function lc((a,t)) to calculate the cost of each step of movement. Among the multiple alignment results calculated, the alignment with the smallest sum of cost values ​​is the optimal alignment.

[0083] For any movement (a, t)∈γ in the alignment, the present invention adopts the following cost function:

[0084] ①(a,t) is log shift, then lc((a,t))=1;

[0085] ②(a,t) is the movement of the model, then lc((a,t))=1;

[0086] ③(a, t) is synchronous movement, then lc((a, t)) = 0;

[0087] ④(a,t) is an illegal move, then lc((a,t))=+∞.

[0088] 2. After the above introduction to the definitions of various parameters described in the present invention, the following mainly introduces a clinical pathway variation detection method based on Petri net synchronous synthesis operation of the present invention.

[0089] Before performing the calculation of synchronous synthesis model alignment, it is necessary to screen the traces in the diagnosis and treatment log, remove noise, and retain valid traces. The following is a detailed description of the clinical pathway variation detection method of the present invention in combination with Section 2.1 (introducing the principle and method of screening valid traces) and Section 2.2 (introducing the method of synchronous synthesis of traces and process models).

[0090] 2.1. Screening of valid traces

[0091] There are generally two possibilities for the traces of event logs with sequence deviation problems: one is that the trace itself is noise; the other is that the activities that could have occurred in parallel in the system did not use the correct parallel representation form, but adopted a sequential relationship or a selection relationship. In clinical pathways, the clinical pathway execution standards must be strictly followed, and most activities are sequential or selection relationships. Therefore, the present invention treats traces with sequence deviation as noise.

[0092] In Petri nets, when a transition can occur before other transitions, there is a precedence relationship between the transition and the subsequent transitions, which is indicated by the symbol "→". When a transition cannot occur before other transitions, there is no precedence relationship between the transition and the other transitions, which is indicated by the symbol "#". For the possible sequence deviation traces in the diagnosis and treatment log, the present invention introduces the concept of precedence relationship matrix.

[0093] Definition 5: Pre-order relation matrix

[0094] Let A be an activity set. N=(P,T;F,α,m i ,m f ) is a Petri net. The pre-order relation matrix is ​​a |T|×|T| matrix, whose row and column labels are the activities mapped by the transitions in the set T, denoted as M:{α(t)|t∈T}×{α(t)|t∈T}→{"→","#"}. The elements of the matrix are as follows:

[0095] ①Yes and like Make Then M[α(t i )][α(t j)]="→", denoted as α(t i )→α(t j ), where 1≤i,j≤n.

[0096] ②Yes and like Make Then M[α(t i )][α(t j )]="#", denoted as α(t i )#α(t j ), where 1≤i,j≤n.

[0097] According to the above rules, we can get N pm The precedence relation matrix of is shown in formula (1).

[0098]

[0099] To determine whether a trace is noise, it is necessary to check all activity pairs in the trace one by one, and compare the order relationship between the currently checked activity and its successor activity with the order relationship in the precedence relationship matrix. If they do not match, the check is terminated, the trace is designated as noise, and removed from the log.

[0100] In order to preserve the information of parallel activities in the "current activity" compared with the "successor activity", it is necessary to set a parallel activity set S as the "current activity set" for comparison with the "successor activity". Before starting the traversal, put the first activity in the trace into S. Let the current activity in the traversal be activity cur, and the successor activity of activity cur in the trace be activity post. The rules are as follows:

[0101] ① If there is one or more activities in the set S whose precedence relation matrix allows the occurrence time to be earlier than the activity post and vice versa, delete these activities from the set and then put the activity post into the set S;

[0102] ② If all activities in set S form a parallel relationship with activity post, only the activity post is placed in set S without deleting any elements;

[0103] ③ If there is any activity in the set S that has a sequence deviation relationship with the activity post, then the trace is regarded as noise and can be discarded directly.

[0104] The screening algorithm is shown in Algorithm 1.

[0105] Algorithm 1 filters the valid traces in the log.

[0106] Input: diagnosis and treatment log L, precedence relation matrix M of process model N;

[0107] Output: Screened diagnosis and treatment logs.

[0108]

[0109]

[0110] To filter the logs, we need to traverse the entire set of diagnosis and treatment logs L and then step by step check the activities in the sequence σ, so two nested loops are required. The time complexity of the algorithm is O(n 2 ). The value of n is affected by the size of the medical log L and the length of the sequence σ. The space complexity of this algorithm is O(n).

[0111] 2.2 Synchronous synthesis of trace and process models

[0112] The so-called synchronous synthesis, that is, for a given trace and its corresponding process model, firstly, the trace needs to be converted into a log model in the form of Petri net, and then the two transitions corresponding to the same activity are merged, and the front set and the back set are merged separately. The new model obtained is the synchronous synthesis model of the trace and the process model, such as Figure 3 shown.

[0113] Figure 3 The log transitions and model transitions with the same label x in can construct synchronous moves. For other non-synchronous moves, the names of the transitions should be modified according to the log transitions and model transitions in Definition 4. The synchronous synthesis algorithm is shown in Algorithm 2.

[0114] Algorithm 2 Synchronous synthesis model generation algorithm.

[0115] Input: Process model N 1 , log model N 2 ;

[0116] Output: Synchronous synthesis model N 3 .

[0117]

[0118]

[0119] The algorithm transforms the process model N 1 With log model N 2 The transitions, places, and flow relations are synthesized separately. Each of the three only requires one loop, so the time complexity is O(n). The value of n is affected by the number of transitions, places, and flow relations. The space complexity of the algorithm is O(n).

[0120] For example Figure 1 The process model N shown pm , assuming a trace σlm =<a,f,b,e> , then the trace σ lm The converted log model is as follows Figure 4 As shown, through Algorithm 2, the log model N lm And process model N pm The calculated synchronous synthesis model N sc like Figure 5 shown.

[0121] 3. Alignment of Synchronous Synthetic Models

[0122] 3.1. Computing the reachability graph of the synchronous synthesis model

[0123] Since alignment is a sequence of moves, and transitions occur with changes in states, the reachability graph can clearly express the changes in various states in the model and the transitions caused by them. Moreover, through the cost function lc((a, t)) introduced in Definition 4, a weight can be assigned to each arc of the reachability graph, representing the cost caused by the transition. The problem of calculating the optimal alignment is converted into the problem of solving the shortest path of a directed weighted graph. Therefore, we need to calculate N sc The reachable graph of N. sc The reachability graph TS sc like Figure 6 shown.

[0124] From Definition 4, we know that: (t 1 ′,t 1 ) is synchronous movement, so lc((t 1 ′,t 1 ))=0;(>>,t 3 ) is the model movement, so lc((>>,t 3 ))=1;(t 2 ′,>>) is log movement, so lc((t 2 ′,>>))=1; By analogy, we can know Figure 6 The cost values ​​of the other transitions are: lc((t 3 ′,t 2 ))=0,lc((>>,t 4 ))=1,lc((t 4 ′,t 5 ))=0.

[0125] 3.2. Finding the Optimal Alignment

[0126] There are many kinds of intelligent search algorithms, which have excellent processing performance for complex problems. *The algorithm is one of the most effective methods in direct search algorithms. When solving the shortest path, its efficiency increases with the accuracy of the heuristic function, that is, the degree of closeness between the estimated value and the actual value, and the code is simple to implement. Since the estimated value calculated by the cost function mentioned in the present invention is basically equal to the actual value, the calculation efficiency can be maximized. Therefore, the present invention is based on A * The algorithm searches for the optimal alignment.

[0127] For the reachable graph TS obtained in Section 3.1 sc , using A * The algorithm calculates the optimal alignment, as shown in Algorithm 3.

[0128] Algorithm 3 uses A * The algorithm finds the optimal alignment based on the reachability graph of the synchronous synthesis model.

[0129] Input: reachability graph TS of synchronous synthesis model N;

[0130] Output: trace σ and process model N 1 An optimal alignment of γ.

[0131]

[0132]

[0133] Since the cost value estimated by the lc((a, t)) cost valuation function used in the present invention is substantially equal to the actual value, A can be achieved. * The efficiency upper limit of the algorithm, considering that the algorithm uses a priority queue, assuming that it uses quick sort, its time complexity is n is related to the number of reachable identifiers in the reachable graph. The number of iterations and the reordering of the priority queue at each iteration are affected by it, so the time complexity is The space complexity is O(n).

[0134] TS sc Take as an example, the optimal alignment calculation result is shown in formula (2).

[0135]

[0136] Algorithm 3 finds the optimal alignment by calculating the minimum cost sum of the transition-induced sequence. At the implementation level, it is not necessary to generate a reachable graph of the synchronous synthesis model. The synchronous synthesis model can be directly used as input as the search space, which simplifies the solution steps, saves processing time, and improves computational efficiency.

[0137] 4. Experiment and evaluate the above-mentioned clinical pathway variation detection method based on Petri net synchronous synthesis operation.

[0138] In order to further illustrate the effectiveness of the computational alignment method proposed in the present invention and evaluate its computational efficiency, the present invention proposes to use the diagnosis and treatment log generated by the actual clinical pathway implementation as an example to study the following issues:

[0139] (1) How efficient is the algorithm proposed in this invention? How does it compare with other algorithms?

[0140] (2) Can the algorithm proposed in this invention obtain correct mutation detection results?

[0141] For a given process model, diagnosis and treatment log, and cost function, the traditional method for calculating the optimal alignment is A proposed by Adriansyah et al. * Alignment algorithm. The present invention adopts A * The algorithm calculates the optimal alignment based on the synchronous synthesis model. Theoretically, compared with the former, it reduces the number of transitions and arcs of the model, so the number of reachable identifiers is also smaller. The efficiency when calculating the optimal alignment should be better than the former.

[0142] Next, we will conduct an experimental verification using the clinical pathway of cardiovascular disease under the COVID-19 pandemic as an example. The process model was manually established based on the diagnosis and treatment record log of a tertiary hospital in Qingdao. Figure 7 As shown in the figure, the corresponding relationship between activities and transitions is shown in Table 2.

[0143] 4.1 Experimental Setup

[0144] The main work of this experiment is to compare the search space scale and calculation alignment efficiency of the method proposed by the present invention with that of the existing methods. All traces in the diagnosis and treatment log are classified according to the ratio of the number of deviations contained in the trace length, from 0% to 30%, with each 5% as a level. Each trace is respectively aligned using the alignment method proposed by the present invention and A * The alignment algorithm calculates the optimal alignment, repeats it 10 times, and calculates the average time and model scale parameters, namely the number of transitions, the number of arc relationships, and the number of nodes in the queue, in order to fully compare the differences between the two under the premise of controlling the variables. The results are as follows: Figure 8-Figure 11 As shown. Among them, Figure 8 , Fig. 9 The figure shows the comparison of the transition number and arc relationship number of the models using the two methods. Fig.10 Comparison of the average number of nodes entering the queue for the two models. Fig.11 Comparison of the average time required to calculate the optimal alignment for the two methods.

[0145] Table 2 Activity and transition correspondence table

[0146]

[0147] 4.2 Experimental Environment

[0148] The experimental code uses Java language. Table 3 shows the hardware and software configuration of the operating platform.

[0149] Table 3 Experimental environment

[0150]

[0151] 4.3 Algorithm superiority verification

[0152] A * The model adopted by the alignment algorithm is a product model, and the method described in the present invention adopts a synchronous synthesis model. The two models will not affect the total number of places during the generation process, but mainly affect the number of transitions and arcs. Figure 8 Comparing the number of transitions of the two models, the synchronous synthesis model has about 37 transitions, and the product model has about 54 transitions. The transitions correspond to actual activities. The synchronous synthesis model reduces some unnecessary transitions, which greatly reduces the size of the model. Fig. 9 The number of arcs of the two models is compared. The number of arcs of the synchronous synthesis model is about 92, and the number of arcs of the product model is about 125. The arc is the flow relationship defined in Definition 2, which represents the association relationship between activities. While reducing transitions, the synchronous synthesis model also reduces some unnecessary flow relationships, thereby reducing the complexity of the search space.

[0153] Fig.10 The average number of nodes in the queue of the two models in the process of calculating the optimal alignment is compared, that is, the number of reachable identifiers in the reachable graph. The number of reachable identifiers of the synchronous synthesis model is about 47, and the number of reachable identifiers of the product model is about 72. The number of reachable identifiers affects A * The number of loops required by the algorithm to find the optimal alignment and the sorting efficiency are shown in Algorithm 3. Figure 8 and Fig. 9 In the , synchronous synthesis model reduces unnecessary transitions and arc relationships compared to the product model, making the reachable graph closer to the actual situation, essentially reducing unnecessary reachable states.

[0154] Fig.11 The average time consumption of the two methods in calculating the optimal alignment is compared. For the same clinical pathway model and the diagnosis and treatment log generated by it, the average time consumption of the algorithm proposed in this invention is about 150ms, while A * The average time taken by the alignment algorithm is about 245ms, which saves about 40% of the time required to calculate the optimal alignment. It is better than A at every noise level. * The alignment algorithm does the job of computing the optimal alignment more quickly.

[0155] There are two main reasons why the method proposed in the present invention is more efficient: first, the method of the present invention pre-processes the diagnosis and treatment logs to remove noise and avoid interference from invalid traces; second, the synchronous synthesis model has the advantage of scale itself, because A* The time complexity of the algorithm will increase with the increase of the number of reachable identifiers of the reachable graph, and the number of reachable identifiers of the reachable graph will increase with the increase of the number of transitions and arcs in the model. Therefore, even if the number of transitions and arcs decreases by less than 30%, the number of nodes entering the queue is still reduced by about 34%, and the time spent on calculating the optimal alignment is reduced by about 40%.

[0156] In summary, the synchronous synthesis model used in the present invention has lower complexity and higher computational efficiency. When calculating the optimal alignment, since the complexity of the reachable graph will increase geometrically with the complexity of the model, the selection of the synchronous synthesis model can greatly improve the alignment efficiency without affecting the variation detection results. For possible noise, the diagnosis and treatment logs can be screened in advance.

[0157] 4.4 Algorithm Validity Verification

[0158] After completing the calculation of the optimal alignment, the optimal alignment result in the form of formula (2) is obtained. From the formula, it can be clearly seen whether the activities in the patient's diagnosis and treatment process are normal, whether there are missing or redundant activities, and medical staff can master the entire clinical pathway diagnosis and treatment process. The following is an example in the diagnosis and treatment log.

[0159] Trace σ = <…, exclude Cov-19 infection, …, CCU thrombolysis, transfer to isolation ward, emergency patient thrombolysis and recanalization, …>, the omitted part is the synchronous transition, and the corresponding relationship between the activities and transitions of this trace is shown in Table 4.

[0160] Table 4 Correspondence between some activities and transitions of trace σ

[0161]

[0162] According to Table 4, the trace σ can be expressed as <…,a,…,b,c,d,…>, and its optimal alignment result is shown in formula (3).

[0163]

[0164] According to the optimal alignment result of formula (3), it can be seen that the patient was initially excluded from COVID-19 infection, but was subsequently transferred to an isolation ward. There may be many reasons for this, such as the worsening of the epidemic, the need to strengthen epidemic prevention and control for high-risk patients with low immunity, or the patient came into contact with outside visitors during treatment, or the patient concealed his personal travel situation at the beginning of diagnosis, etc. The specific situation requires medical staff to analyze it in combination with the actual situation, which cannot be judged only from the log. *The alignment algorithm is an algorithm that has been widely recognized and used for a long time, and its effectiveness is beyond doubt. Therefore, the present invention compares the optimal alignments calculated by the two algorithms. If the results are the same, the algorithm proposed by the present invention is considered to be effective. If the results are different, it is considered to be invalid.

[0165] Since the result of the optimal alignment is a (α 2 (t x ′),(α 1 (t y ),t y )) form of nodes, where t x ′ corresponds to the change of the log model, α 2 (t x ′) is the name of the activity corresponding to the transition in the log model, t y Corresponding to the change of process model, α 1 (t y ) is the activity name corresponding to the transition in the process model. Therefore, after completing the optimal alignment respectively, the node contents contained in the optimal alignment results obtained by the two algorithms are compared one by one in order to check whether the process model label, log model label, and activity name are exactly the same. According to the inspection result, it can be judged whether the method proposed by the present invention is effective.

[0166] When performing the computational alignment, the two optimal alignment results for each trace in the diagnosis and treatment log were compared, and the inspection results were all the same, which can prove that the algorithm of the present invention is effective in variant detection.

[0167] 5. The method proposed in the present invention can be applied to a computer. Specifically, a clinical pathway variation detection method based on Petri net synchronous synthesis operation is applied to a program running on a computer processor.

[0168] Further preferably, the above-mentioned computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, a clinical pathway variation detection method based on Petri net synchronous synthesis operation as described above is implemented.

[0169] Further preferably, the above-mentioned computer-readable storage medium has a computer program, and when the computer program is executed by the processor, the processor executes the above-mentioned clinical pathway variation detection method based on Petri net synchronous synthesis operation.

[0170] In summary, the present invention performs variation detection on clinical pathways, which is an important means to standardize and study the diagnosis and treatment process and optimize clinical pathways. *The alignment algorithm is a widely applicable compliance checking algorithm that can multiply the trace with the process model. It has made a qualitative leap compared to the previous manual forward-looking method. However, due to the high complexity of the model, the efficiency is still unsatisfactory when calculating the optimal alignment. Therefore, the present invention proposes an optimization algorithm.

[0171] The present invention first pre-processes the diagnosis and treatment log to save time for subsequent work. Then, considering that the complexity of the reachable graph of the Petri net will increase with the increase of the scale of the Petri net, and alignment is actually based on the reachable graph to find the shortest path, reducing unnecessary changes in the synthetic model, even if the number of changes is only slightly reduced, the alignment efficiency can still be greatly improved. Therefore, the present invention adopts a synchronous synthetic model for alignment calculation, which greatly reduces the time consumed by alignment.

[0172] Finally, the present invention conducted an experimental analysis using the clinical pathway of cardiovascular diseases under the COVID-19 pandemic as an example, proving that the variation detection algorithm proposed in the present invention has the advantage of higher efficiency than the traditional algorithm, and can greatly shorten the time consumed in the variation detection process. In subsequent work, this method can be applied to online variation detection. Online variation detection can promptly discover and remind patients of variations in the currently completed diagnosis and treatment process, or estimate possible subsequent variations, and provide a basis for medical personnel to conduct manual intervention to avoid adverse events. Online variation detection has higher time requirements, which can better reflect the efficiency advantages of the method proposed in the present invention.

[0173] Parts not described in the present invention can be implemented by referring to the existing technology.

[0174] It should be noted that any equivalent manner or obvious variation manner made by those skilled in the art under the guidance of this specification should be within the protection scope of the present invention.

Claims

1. A clinical pathway variation detection method based on Petri net synchronous synthesis operation, It is characterized in that The following steps are included in sequence: S1. Screen the traces in the clinical pathway diagnosis and treatment log, remove noise, and retain valid traces; S2. Establishing a new synthesis model based on the effective trace and its corresponding process model, wherein the new synthesis model reflects the deviation between the diagnosis and treatment log and the clinical pathway; S3, using A whose time complexity is positively correlated with the complexity of the selected synthesis model * The algorithm finds the optimal alignment; In step S1, when screening the traces in the clinical pathway diagnosis and treatment log, all activities in the trace are checked one by one, and the order relationship between the currently checked activity and its subsequent activities is compared with the order relationship in the precedence relationship matrix. If they do not match, the check is terminated, and the trace is designated as noise and removed from the log; In step S1, a parallel activity set S is set. Before starting the traversal, the first activity in the trace is placed in S. The current activity of the traversal is set as activity cur, and the successor activity of activity cur in the trace is set as activity post. The rules are as follows: S11. If there is one or more activities in the set S whose precedence relation matrix allows the occurrence time to be earlier than the activity post and vice versa, delete these activities from the set and then put the activity post into the set S; S12. If all activities in set S form a parallel relationship with activity post, only activity post is placed in set S without deleting any elements; S13, if there is any activity in the set S that has a sequence deviation relationship with the activity post, then the trace is regarded as noise and can be directly discarded; In step S2, when establishing a new synthetic model, first convert the effective trace into a log model in the form of a Petri net, then merge two transitions corresponding to the same activity, and merge the front set and the back set respectively, so as to obtain a new synthetic model; In step S3, the reachable graph of the new synthetic model is first calculated, and then based on A * The algorithm searches for the optimal alignment.

2. A clinical pathway variation detection method based on Petri net synchronous synthesis operation according to claim 1, Features: The alignment result is evaluated by introducing a cost function lc((a, t)), which is used to calculate the cost of each step of movement. Among the multiple alignment results calculated, the alignment with the smallest sum of cost values ​​is the optimal alignment; For any movement (a, t)∈γ in the alignment, the cost function is as follows: S31, (a, t) is log movement, then lc((a, t)) = 1; S32, (a, t) is the model movement, then lc((a, t)) = 1; S33, (a, t) is synchronous movement, then lc((a, t)) = 0; S34, (a, t) is an illegal move, then lc((a, t)) = +∞.

3. A clinical pathway variation detection method based on Petri net synchronous synthesis operation according to claim 1, Features: For any movement (a, t)∈γ in the alignment, there are four conditions: S35. If a∈A, t=>>, (a, t) is called a log move; S36. If a=>>, t∈T, (a, t) is called model movement; S37. If a∈A, t∈T, (a, t) is called synchronous movement; S38, the rest are called illegal movements; Among them, A is an activity set.

4. Application of a clinical pathway variation detection method based on Petri net synchronous synthesis operation according to any one of claims 1 to 3 in a computer, Features: The application is to apply the clinical pathway variation detection method based on Petri net synchronous synthesis operation to a program running on a processor of a computer.

5. A computer, Features: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a clinical pathway variation detection method based on Petri net synchronous synthesis operation as claimed in any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, Features: The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor executes a clinical pathway variation detection method based on Petri net synchronous synthesis operation as described in any one of claims 1 to 3.

Images