Chronic disease comorbidity prediction method and system based on time series disease network

Abstract

The present invention belongs to the field of medical data processing technology, and discloses a chronic disease comorbidity prediction method and system based on a time series disease network. The present invention constructs a node representation method of a time series disease network to mine the topological structure information of the time series disease network and obtain accurate disease node representation vectors. On this basis, a time encoding module and a time-aware long-short-term memory network capable of perceiving time information are constructed, and modeling is performed for irregular hospitalization time intervals to capture the impact of time factors on disease evolution, which can more accurately predict the development trend of the disease and the risks that patients may face in the future.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to a method and system for predicting chronic disease comorbidity based on a time series disease network. Background Art

[0002] With the aging of the population and the complex evolution of the disease spectrum, chronic diseases and their comorbidities have become a major public health problem that threatens people's lives and health, and has brought heavy service supply and economic burdens to the medical and health system. Compared with a single chronic disease, comorbidities not only significantly increase the risk of physical functional limitations, disability, and even death in patients, but also increase the complexity and difficulty of diagnosis and treatment, resulting in a significant increase in medical service utilization and family health expenditures, which has brought severe challenges to the traditional medical service management model of prevention, diagnosis, and treatment of a single disease. With the convergence of massive health and medical big data and the development of emerging technologies such as artificial intelligence, research on data-driven risk prediction of chronic diseases and their comorbidities has made certain progress. However, existing research has the following problems: (1) Research on risk prediction of chronic diseases and their comorbidities mainly focuses on mining individual patient characteristics, and there is still a gap in how to effectively use the characteristics and laws of disease evolution to support prediction. (2) Existing methods treat multiple visits of individual patients in longitudinal medical data as medical events of equal intervals, ignoring the irregularity of the time interval distribution and the health status of the patient revealed.

[0003] In view of this, the present invention is proposed. Summary of the invention

[0004] The present invention aims to solve at least one of the above technical problems and provides a method and system for predicting chronic disease comorbidities based on a time series disease network.

[0005] To achieve the above objectives, the present invention provides the following technical solutions.

[0006] In a first aspect, the present invention provides a method for predicting chronic disease comorbidity based on a temporal disease network, comprising the following steps:

[0007] Obtain patients' historical hospitalization records, including each patient's demographic characteristics, admission and discharge times, and disease diagnosis;

[0008] constructing a temporal disease network based on an analysis data set randomly divided from the historical hospitalization records;

[0009] Calculating a disease node vector of a temporal disease network using a reachability-based attention mechanism, wherein the disease node vector contains topological structure information of the temporal disease network;

[0010] The disease node vector, demographic characteristics and unique-hot encoding of each patient’s hospitalization record are integrated according to the view-level attention mechanism to obtain the hospitalization record vector of each patient;

[0011] Merging the hospitalization record vector and a first time interval as a first input, wherein the first time interval is a time interval between each hospitalization record of each patient and the last hospitalization record of the patient;

[0012] Inputting a first input and a second time interval as input data into a time-aware long short-term memory network model to obtain a final hidden state of the time-aware long short-term memory network, wherein the second time interval is a time interval between two adjacent hospitalization records of each patient;

[0013] The final hidden state is subjected to a Softmax operation to obtain the risk of patients having chronic comorbidities.

[0014] Preferably, the method of fusing the hospitalization record vector and the first time interval as the first input is:

[0015] encoding a first time interval;

[0016] Add the encoded first time interval to the hospitalization record vector.

[0017] Preferably, the formula for encoding the first time interval is: Among them, W1, W2, b1 and b2 are trainable parameters, a is a hyperparameter, and ΔT is the first time interval.

[0018] Preferably, the second time interval is modeled by adjusting the cell memory of the time-aware long short-term memory network, and the formula is:

[0019]

[0020] Among them C r-1 It is the cell memory from the previous time, W d and b d is a trainable parameter, are two intermediate variables generated during the calculation process, Δt r-1 is the second time interval and tanh is the activation function.

[0021] Preferably, the method of integrating the disease node vector, demographic characteristics and one-hot encoding of the disease of each patient's hospitalization record for each time according to the view-level attention mechanism to obtain the vector of each patient's hospitalization record for each time is as follows: for each patient's hospitalization record for each time, all disease node vectors in the hospitalization record are integrated into a unified disease vector according to the disease code attention mechanism; and the unified disease vector, demographic characteristics and one-hot encoding of the disease are integrated according to the view-level attention mechanism.

[0022] Preferably, the unified disease vector Z e It is expressed as:

[0023] Where Ρ(r) represents the set of all diseases in the hospitalization record r; a, W CA and b CA is a trainable parameter; tanh is an activation function; T represents vector transposition; e i =Attention(x i ).

[0024] Preferably, the hospitalization record vector is expressed as:

[0025] Where K is the demographic characteristic Z d 、The unique hot encoding Z of the disease b and the unified disease vector Z e The subscripts of are the set, S is the set of analysis data sets, q, W VA and b VA is a trainable parameter, tanh is the activation function, p is the patient, and T represents the vector transpose.

[0026] Preferably, the risk of a patient having a chronic disease comorbidity is expressed as: in, is the predicted probability, W λ and b λ is a trainable parameter, h λ is the final hidden state.

[0027] Preferably, a randomly divided dataset from the patient's historical hospitalization records is used to train and verify the time-aware long short-term memory network model, and the loss function is: Where N represents the number of patients in the training set, is the chronic disease comorbidity risk prediction label for patient p, and T represents the vector transpose.

[0028] In a second aspect, the present invention provides a chronic disease comorbidity prediction system based on a time series disease network, comprising:

[0029] The acquisition module is used to obtain the patient's historical hospitalization records, including each patient's demographic characteristics, admission and discharge time, and disease diagnosis;

[0030] A construction module, used for constructing a temporal disease network based on the analysis data set randomly divided in the historical hospitalization records;

[0031] A calculation module, used for calculating a disease node vector of a temporal disease network using an attention mechanism based on reachability, wherein the disease node vector contains topological structure information of the temporal disease network;

[0032] An integration module is used to integrate the disease node vector, demographic characteristics, and one-hot encoding of each patient's hospitalization record according to the view-level attention mechanism to obtain the hospitalization record vector of each patient;

[0033] A fusion module, used for fusing the hospitalization record vector with a first time interval as a first input, wherein the first time interval is a time interval between each hospitalization record of each patient and the last hospitalization record;

[0034] An output module, configured to input a first input and a second time interval as input data into a time-aware long short-term memory network model, and output a final hidden state of the time-aware long short-term memory network, wherein the second time interval is a time interval between two adjacent hospitalization records of each patient;

[0035] The prediction module is used to perform a Softmax operation on the final hidden state to predict the risk of patients with chronic comorbidities.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention constructs a node representation method for a temporal disease network to mine the topological structure information of the temporal disease network and obtain accurate disease node representation vectors. On this basis, a time encoding module and a time-aware long short-term memory network that can perceive time information are constructed. Irregular hospitalization time intervals are modeled to capture the impact of time factors on disease evolution, which can more accurately predict the development trend of the disease and the risks that patients may face in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of a process for predicting chronic disease comorbidities based on a time series disease network provided in an embodiment of the present invention;

[0039] Figure 2 A schematic diagram for representing nodes in a temporal disease network;

[0040] Figure 3A schematic diagram of the structure of a chronic disease comorbidity prediction system based on a time series disease network provided in an embodiment of the present invention;

[0041] Figure 4 FIG. 1 is a schematic block diagram of an exemplary electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] The research population of the present invention is: people suffering from a certain chronic disease, and the specific chronic disease is defined as the index disease. The goal of the present invention is: based on the λ hospitalization history records of patients with the index disease, predict whether they will have another chronic disease comorbidity in the next τ years, and the above chronic disease comorbidity is defined as the target disease.

[0044] In view of the above objectives, the first embodiment of the present invention provides a method for predicting chronic disease comorbidity based on a time series disease network. Figure 1 A flowchart of a chronic disease comorbidity prediction method based on a time series disease network provided in an embodiment of the present application.

[0045] like Figure 1 As shown in the figure, the chronic disease comorbidity prediction method based on time series disease network specifically includes the following steps:

[0046] S101, obtain the patient's historical hospitalization records, including each patient's demographic characteristics, admission and discharge time, and disease diagnosis.

[0047] The patient can be all chronic disease patients in a certain area, a certain city, a certain province, a certain country, etc. The larger the number of patients, the more accurate the final prediction result. Generally speaking, the patient's hospitalization record comes from the patient's medical record homepage data, which records the patient's hospitalization information, including demographic characteristics (such as patient ID number, gender, age, etc.), admission and discharge time, and disease diagnosis. The personal information such as the patient's ID number involved in the present invention is obtained through legal channels, and the clear authorization and consent of the information subject have been obtained. The data collection and processing process strictly abides by the requirements of relevant laws and regulations to ensure that personal privacy rights and interests are not infringed.

[0048] Using the patient ID number as the index and the discharge time as the order, the multiple hospitalization records of each patient are represented as a two-dimensional variable-length array. iFor example, the patient has m hospitalization records, each of which includes admission and discharge time, demographic characteristics, and 16 disease diagnoses.

[0049] S102, constructing a time series disease network according to the randomly divided analysis data set in the historical hospitalization records.

[0050] A certain proportion of the hospitalization records obtained in step S101 is randomly divided as an analysis data set to construct a time series disease network. The proportion can be selected according to actual conditions without special restrictions, for example, 60%, 70%, etc.

[0051] The discharge time of the hospitalization record where the disease first appeared is defined as the diagnosis time of the disease. All directed disease pairs with a diagnosis time span of H years (such as D1→D2, D1 is a precursor disease, and D2 is a subsequent disease) in the entire analysis data set are counted, and the directed disease pairs are merged into a temporal disease network, where H is determined according to the onset cycle of different chronic diseases, etc., without special restrictions, for example, H can be 3, 5, 6, 10, etc. The method for constructing a temporal disease network in the embodiment of the present invention is a method disclosed in the prior art and does not need to be specially limited. Specifically, the method for constructing a temporal disease network in the embodiment of the present invention may include the following steps:

[0052] S102-1, calculating the Relative Ration (RR) value of each directed disease pair, and performing statistical tests to find out the disease pairs whose occurrence probability is significantly higher than the random situation.

[0053] By calculating the RR value, it is determined whether patients with precursor diseases are more likely to develop subsequent diseases than patients in the random group. Specifically, taking the directed disease pair D1→D2 as an example, patients with D1 disease are defined as the case group, and a control individual is matched and randomly selected for each case in a 1:1 ratio with the same gender and age group and a discharge time interval of less than one week as matching conditions to form the control group. The time when each patient in the case group is diagnosed with D1 disease is defined as the index date of the patient. Accordingly, the index date of the patients in the control group is the same as that of the matched case group subjects.

[0054] It should be noted that the number of patients in the analysis data set is huge, and the case group consists of patients with disease D1. According to the above matching conditions, each patient in the case group can be matched with multiple candidate individuals that meet the above conditions, and one of the multiple candidate individuals is randomly selected as a control individual. Repeat this random selection process M times. Since the control individuals randomly selected each time are random, M control groups can be formed, and finally 1 case group and M control groups are obtained.

[0055] The calculation formula of RR value is: Among them, C exposed represents the number of patients in the case group who developed disease D2 within H years after the index date; M represents the number of controls; C i represents the number of people in the ith control group who develop disease D2 within H years after the index date.

[0056] The formula for calculating the P value is:

[0057] S102-2, determine the direction of disease progression by binomial test. In particular, when the RR of D1→D2 and D2→D1 are both greater than 1 and statistically significant, at most only one direction is confirmed as the final progression direction.

[0058] S102-3, obtain linear disease trajectories and construct a temporal disease network.

[0059] All the directed disease pairs obtained above are merged into a linear disease trajectory with a disease number of β, where β is a positive integer greater than 1. The more diseases β there are in the disease trajectory, the fewer patients will have this disease trajectory. When the number of patients is too small (close to 0), the disease trajectory is meaningless. Therefore, the value of β can be selected according to the actual situation of the corresponding chronic disease and comorbidity, without special restrictions. Specifically, taking β = 4 as an example, the directed disease pairs are sorted from large to small according to the number of patients, and then scanned from front to back to calculate the number of occurrences of directed disease pairs with overlapping diseases in turn; if the number of occurrences is greater than 0, the two disease pairs are merged into a linear disease trajectory with a disease number of 3; otherwise, the scanning is stopped. Similarly, the linear disease trajectory with a disease number of 3 is further merged into a linear disease trajectory with a disease number of 4. Finally, the linear disease trajectory with a disease number of 4 is merged according to the disease nodes to obtain a temporal disease network.

[0060] S103, using the reachability-based attention mechanism to calculate the disease node vector of the temporal disease network, wherein the disease node vector contains the topological structure information of the temporal disease network.

[0061] The key to this step is to use the temporal disease network as input, and use the method based on the reachability attention mechanism to output the disease node vector representation. The purpose of this step is to mine the topological structure information of the temporal disease network and obtain the disease node representation vector more accurately. The specific calculation method of the disease node network, including the specific method based on the reachability attention mechanism, does not need to be specially limited, as long as the disease node vector containing the topological structure information of the temporal disease network can be obtained.

[0062] In a specific embodiment, Figure 2As shown, the temporal disease network obtained in S102 is a directed acyclic graph, defined as G = (V, E), where V is a node set, The reachability between nodes is represented by ≤. Specifically, if u,v∈V, then u≤v means that there is a path from u to v in G, and u is the predecessor node of v and v is the successor node of u. Furthermore, (u,v)∈≤ k Indicates that u can reach v through a path of length less than k. In a temporal disease network, if the node with zero in-degree is defined as the initial node, the depth of the node v∈V is defined as follows:

[0063]

[0064] In a time series disease network, each disease node has special position information, that is, the depth of the node in the graph. The embodiment of the present invention includes the depth information of the node into the node representation calculation through directed acyclic graph position coding. Specifically, the directed acyclic graph position coding calculation formula is as follows:

[0065]

[0066] Among them, d is the dimension of the node vector representation, and i is the index of the vector dimension.

[0067] In a preferred embodiment, in order to fully explore the topological structure of the temporal disease network, a disease node vector representation with good representation ability is obtained through an attention mechanism based on reachability. Specifically, the calculation formula of the attention mechanism based on reachability is as follows:

[0068]

[0069] Wherein, k is a hyperparameter, which can be determined based on the topological properties of the temporal disease network. Optionally, k = ∞; Ν k (v)={(u,v)∈≤ k}∪{(v,u)∈≤ k}; <γ,η> represents the dot product of γ and η; x v and x u is the initialization vector representation of nodes v and u, which can be encoded using one-hot encoding; W Q and W k is the trainable parameter matrix; d K is the dimension of the node feature vector; f(x u )=x u W V ; W V is the trainable parameter matrix; PE v PE u PE wBoth represent a vector, PE(v, 2i) represents the even dimension of this vector, and PE(v, 2i+1) represents the odd dimension of this vector.

[0070] S104, integrating the disease node vector, demographic characteristics and unique-hot encoding of the disease of each patient's hospitalization record each time according to the view-level attention mechanism to obtain the hospitalization record vector of each patient each time.

[0071] The implementation of this step includes:

[0072] S104-1, for each hospitalization record of each patient, all disease node vectors in the hospitalization record are integrated into a unified disease vector according to the disease code attention mechanism.

[0073] In order to obtain the vector representation of the hospitalization record, we first integrate all the disease node vectors in each hospitalization record of each patient obtained in S102 through disease coding attention to obtain a unified disease vector Z e For example, if a patient has 5 hospitalization records, then multiple disease diagnoses in each hospitalization record of each patient are integrated to obtain 5 unified disease vectors Z e . Z e The calculation formula is:

[0074] Where Ρ(r) represents the set of all diseases in each patient's rth hospitalization record; a, W CA and b CA is a trainable parameter; tanh is an activation function; T represents vector transposition; e i =Attention(x i ).

[0075] S104-2, according to the view-level attention mechanism, the unified disease vector, demographic characteristics and unique-hot encoding of the disease of each patient's hospitalization record are integrated to obtain the hospitalization record vector of each patient.

[0076] Hospitalization record vector Z r The formula is:

[0077] Where K is the demographic characteristic Z d 、The unique hot encoding Z of the disease b and the unified disease vector Z e is a set of subscripts; S is a set of analysis data sets; q, W VA and b VA is a trainable parameter; tanh is the activation function; p is the patient; T represents the vector transpose.

[0078] S105: Merge the hospitalization record vector and a first time interval as a first input, where the first time interval is the time interval between each hospitalization record of each patient and the last hospitalization record.

[0079] Specifically, the time interval ΔT between each hospitalization record and the last hospitalization of each patient is calculated, and ΔT is encoded by the following formula:

[0080] Among them, W1, W2, b1 and b2 are trainable parameters, and a is a hyperparameter.

[0081] The vector representation of hospitalization record r is Z r and ΔT' r Adding, we can get the first input X r .

[0082] For the first time interval, for example, if a patient has four hospitalization records, namely a, b, c, and d, the first time interval includes the time intervals between a and d, b and d, and c and d.

[0083] S106, inputting a first input and a second time interval as input data into a time-aware long short-term memory network model to obtain a final hidden state of the time-aware long short-term memory network, wherein the second time interval is the time interval between two adjacent hospitalization records of each patient.

[0084] For the second time interval, for example, if a patient has four hospitalization records, namely a, b, c, and d, the second time interval includes the time intervals between a and b, b and c, and c and d.

[0085] The first input {X1,X2,...,X r ,...,X λ} and the time interval between two consecutive hospitalization records of each patient {Δt0,...,Δt r ,...,Δt λ-1}Input into the time-aware long short-term memory network and output the final hidden state.

[0086] Specifically, the computational process of the time-aware long short-term memory network is as follows:

[0087] First, calculate the forget gate f of the time-aware LSTM network r , input gate i r and output gate o r , this calculation method is a conventional method in the traditional LSTM model, and the specific formula is as follows:

[0088] f r =σ(W f Xr +U f h r-1 +b f )

[0089] i r =σ(W i X r +U i h r-1 +b i )

[0090] o r =σ(W o X r +U o h r-1 +b o )

[0091] Among them, {W f ,U f ,b f}、{W i ,U i ,b i} and {W o ,U o ,b o} are three sets of trainable parameters; h r-1 is the previous hidden state; σ is the activation function.

[0092] In a preferred embodiment, since the time intervals between two consecutive hospitalizations of a patient are often not exactly the same, the impact of the previous hospitalization on the subsequent hospitalization is also different. Therefore, in order to more accurately reflect the differences in the impacts of different time intervals, the irregular time intervals between hospitalizations are modeled by adjusting the cell memory of the time-aware long short-term memory network, that is, the second time interval is modeled:

[0093]

[0094] Among them, C r-1 It is the cell memory from the previous time, W d and b d is a trainable parameter, are two intermediate variables generated during the calculation process, Δt r-1 is the second time interval and tanh is the activation function.

[0095] Current Cell Memory C r and the hidden state h r It can be updated by:

[0096]

[0097] hr =o r *tanh(C r )

[0098] Among them, W c , U c and b c is a trainable parameter.

[0099] S107, performing a Softmax operation on the final hidden state to obtain the risk of the patient having a chronic disease comorbidity.

[0100] After obtaining the final hidden state h of the time-aware long short-term memory network λ Finally, through the linear layer and the Softmax function, we can get whether the patient has the risk of the target disease: in, is the predicted probability, W λ and b λ is a trainable parameter. In particular, if the patient has the target disease before prediction, then the output is

[0101] The embodiment of the present invention uses a certain proportion of historical hospitalization records as a data set for training and verifying the prediction model, that is, the remaining hospitalization records in the historical hospitalization records except the analysis data set are used as the training and verification prediction model data set. The specific proportion can be selected according to the actual situation, such as 40% of the historical hospitalization records. This part of the data is further divided into a training data set and a test data set, which is determined according to the division ratio commonly used in the field. For example, the data content of the training data set and the test data set can be 8:2. The training data set is used for training data. On the training data set, the model parameters are updated by optimizing the following loss function:

[0102]

[0103] Where N represents the number of patients in the training set, is the chronic disease comorbidity risk prediction label for patient p, and T represents the vector transpose.

[0104] The second embodiment of the present invention provides a chronic disease comorbidity prediction system based on a time series disease network.

[0105] Figure 3 A schematic diagram of the structure of a chronic disease comorbidity prediction system based on a time series disease network provided in an embodiment of the present application.

[0106] like Figure 3As shown, the chronic disease comorbidity prediction system 300 based on the time series disease network includes an acquisition module 301, a construction module 302, a calculation module 303, an integration module 304, a fusion module 305, an output module 306, and a prediction module 307. The functions of each module are described in detail as follows:

[0107] The acquisition module 301 is used to obtain the patient's historical hospitalization records, including each patient's demographic characteristics, admission and discharge time, and disease diagnosis;

[0108] A construction module 302 is used to construct a time series disease network according to the randomly divided analysis data set in the historical hospitalization record;

[0109] A calculation module 303 is used to calculate a disease node vector of a time series disease network, wherein the disease node vector contains topological structure information of the time series disease network;

[0110] An integration module 304 is used to integrate the disease node vector, demographic characteristics and one-hot encoding of the disease of each patient's hospitalization record each time according to the view-level attention mechanism to obtain a hospitalization record vector of each patient each time;

[0111] A fusion module 305 is used to fuse the hospitalization record vector with a first time interval as a first input, where the first time interval is a time interval between each hospitalization record and the last hospitalization of each patient;

[0112] An output module 306 is configured to input a first input and a second time interval as input data into a time-aware long short-term memory network model, and output a final hidden state of the time-aware long short-term memory network, wherein the second time interval is a time interval between two adjacent hospitalization records of each patient;

[0113] The prediction module 307 is used to perform a Softmax operation on the final hidden state to predict the risk of the patient having a chronic disease comorbidity.

[0114] Based on the above embodiments, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the steps in the chronic disease comorbidity prediction method based on a time series disease network in the first aspect of the embodiment are implemented.

[0115] Figure 4A schematic block diagram of an example electronic device 400 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0116] like Figure 4 As shown, the electronic device 400 may include a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. In the RAM 403, various programs and data required for the operation of the device 400 can also be stored. The computing unit 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0117] Multiple components in the device 400 are connected to the I / O interface 405, including: an input unit 406, such as a keyboard, a mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a disk, an optical disk, etc.; and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the device 400 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0118] The computing unit 401 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 401 performs the various methods and processes described above, such as omics data processing methods or model training methods for omics data processing. For example, in some embodiments, the omics data processing method or the model training method for omics data processing may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the omics data processing method or the model training method for omics data processing described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to execute the omics data processing method or the model training method for omics data processing in any other suitable manner (eg, by means of firmware).

[0119] Based on the above embodiments, the present invention further provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the cancer dynamic survival analysis method based on hierarchical attention disclosed in the embodiments of the present invention.

[0120] Based on the above embodiments, the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the cancer dynamic survival analysis method based on hierarchical attention disclosed in the embodiments of the present invention.

[0121] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer readable media do not include electric carrier signals and telecommunication signals.

[0122] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0123] In the several embodiments provided in the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0124] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0125] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0126] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of each embodiment of the present application.

[0127] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A chronic disease comorbidity prediction method based on a time series disease network, characterized in that: The following steps are involved: Obtain the patient's historical hospitalization records, including each patient's demographic characteristics, admission and discharge times; constructing a temporal disease network based on an analysis data set randomly divided from the historical hospitalization records; Calculating a disease node vector of a temporal disease network using a reachability-based attention mechanism, wherein the disease node vector contains topological structure information of the temporal disease network; The disease node vector, demographic characteristics and unique-hot encoding of each patient’s hospitalization record are integrated according to the view-level attention mechanism to obtain the hospitalization record vector of each patient; The first time interval is encoded, and the encoded first time interval is added to the hospitalization record vector and fused into a first input, wherein the first time interval is the time interval between each hospitalization record of each patient and the last hospitalization record; the formula for encoding the first time interval is: Among them, W1, W2, b1 and b2 are trainable parameters, a is a hyperparameter, and ΔT is the first time interval. The second time interval is modeled by adjusting the cell memory of the time-aware long short-term memory network. The second time interval is the time interval between two adjacent hospitalization records of each patient. The formula for modeling the second time interval is: Among them C r-1 It is the cell memory from the previous time, W d and b d is a trainable parameter, are two intermediate variables generated during the calculation process, Δt r-1 is the second time interval, tanh is the activation function; Inputting the first input and the second time interval as input data into the time-aware long short-term memory network model to obtain a final hidden state of the time-aware long short-term memory network; The final hidden state is subjected to a Softmax operation to obtain the risk of patients having chronic comorbidities.

2. The method according to claim 1, characterized in that The method for integrating the disease node vector, demographic characteristics and one-hot encoding of the disease of each patient's hospitalization record for each time according to the view-level attention mechanism to obtain the vector of each patient's hospitalization record for each time is as follows: for each patient's hospitalization record for each time, all disease node vectors in the hospitalization record are integrated into a unified disease vector according to the disease code attention mechanism; and the unified disease vector, demographic characteristics and one-hot encoding of the disease are integrated according to the view-level attention mechanism.

3. The method according to claim 2, characterized in that The unified disease vector Z e It is expressed as: Where Ρ(r) represents the set of all diseases in the hospitalization record r; a, W CA and b CA is a trainable parameter; tanh is an activation function; T represents vector transposition; e i =Attention(x i ).

4. The method according to claim 2 or 3, characterized in that The hospitalization record vector is expressed as: Where K is the demographic characteristic Z d 、The unique hot encoding Z of the disease b and the unified disease vector Z e The subscripts of are the set, S is the set of analysis data sets, q, W VA and b VA is a trainable parameter, tanh is the activation function, p is the patient, and T represents the vector transpose.

5. The method according to claim 1, characterized in that The risk of a patient having a chronic disease comorbidity is expressed as: in, is the predicted probability, W λ and b λ is a trainable parameter, h λ is the final hidden state.

6. The method according to claim 1, characterized in that A randomly divided dataset from the patient hospitalization records is used to train and validate the time-aware long short-term memory network model, and the loss function is: Where N represents the number of patients in the training set, is the predicted probability of chronic disease comorbidity risk of patient p, y p is the chronic disease comorbidity risk prediction label of patient p, and T represents the vector transpose.

7. Chronic disease comorbidity prediction system based on time series disease network, characterized by: Include: The acquisition module is used to obtain the patient's historical hospitalization records, including each patient's demographic characteristics, admission and discharge time, and disease diagnosis; A construction module, used for constructing a temporal disease network based on the analysis data set randomly divided in the historical hospitalization records; A calculation module, used for calculating a disease node vector of a temporal disease network using an attention mechanism based on reachability, wherein the disease node vector contains topological structure information of the temporal disease network; The integration module is used to integrate the disease node vector, demographic characteristics and unique hot encoding of the disease of each hospitalization record of each patient according to the view-level attention mechanism to obtain the hospitalization record vector of each patient each time; the fusion module is used to encode the first time interval, add the encoded first time interval and the hospitalization record vector to fuse as the first input, the first time interval is the time interval between each hospitalization record of each patient and its last hospitalization record; and the second time interval is modeled by adjusting the cell memory of the time-aware long short-term memory network, the second time interval is the time interval between two adjacent hospitalization records of each patient; the formula for encoding the first time interval is: Among them, W1, W2, b1 and b2 are trainable parameters, a is a hyperparameter, ΔT is the first time interval; the formula for modeling the second time interval is: Among them C r-1 It is the cell memory from the previous time, W d and b d is a trainable parameter, are two intermediate variables generated during the calculation process, Δt r-1 is the second time interval, tanh is the activation function; An output module, configured to input the time-aware long short-term memory network model with the first input and the second time interval as input data, and output a final hidden state of the time-aware long short-term memory network; The prediction module is used to perform a Softmax operation on the final hidden state to predict the risk of patients with chronic comorbidities.

Images