Random forest construction method, device and personalized medication regimen design method

Abstract

The application discloses a random forest construction method and device and a personalized medication scheme design method, wherein the method comprises the following steps: when a node of any decision tree is divided, the node is divided according to the following strategy: the joint distribution heterogeneity of a medication scheme variable and a patient medication outcome variable under the condition of a given patient individual characteristic in the divided node is reduced; and a plurality of decision trees are established according to the strategy, and the random forest is generated, so that the technical problem that the existing medication scheme design technology cannot guarantee applicability and accuracy under the condition that a plurality of factors are comprehensively affected and drug efficacy is obviously heterogeneous among patients in the related art can be solved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence in medication design, and in particular to a random forest construction method, apparatus, and personalized medication design method. Background Technology

[0002] Modern medical practice shows that the therapeutic effect of drugs is related to the medication regimen (whether to use medication, combination of medications, timing of administration, and dosage), and is also influenced by genetic factors and individual dominant traits (such as sex, age, and weight). Even with the exact same medication regimen, the effects can vary substantially among different patient subgroups; furthermore, each patient has a different combination of genetic and dominant traits, making them relatively unique, and clinicians need to make choices among different medication regimens.

[0003] Regarding medication regimens, modern medicine has successively progressed through population-based research models (such as randomized controlled trials comparing the average blood pressure-lowering effects of two antihypertensive drugs in different patient groups), stratified models (such as exploring the applicability of different targeted drugs based on epidermal growth factor receptor (EGFR) genotypes), and the exploration of personalized medication regimens. Personalized medication is both an inevitable requirement of the development trend of precision medicine and a challenge to medication regimen design technology.

[0004] Currently, personalized medicine mainly involves the following types of technologies: ① Genetic trait-based bioassay technology. This involves using gene sequencing or diagnostic kits to pre-test genetic traits (such as genotype) that have been proven to affect treatment outcomes, and then tailoring treatment plans to patients with different genetic traits. Advances in this type of technology depend on the rate of laboratory research output, and clinical effects are often evaluated only in small groups of patients. It also tends to overlook factors such as individual dominant traits, and its extrapolation still faces several challenges.

[0005] ② Predictive techniques based on statistics or machine learning. These techniques integrate genetic information and individual characteristics, and use medication information as a predictor of treatment efficacy. Machine learning techniques are currently in a growth phase in research areas such as immunosuppressants, anti-infective drugs, and antiepileptic drugs. However, this approach struggles to provide efficacy estimates for different medication regimens for patients with given characteristics, and when significant heterogeneity in drug efficacy exists among patients, the accuracy and extrapolation of the predictive techniques are difficult to guarantee.

[0006] ③ Reinforcement learning techniques combined with causal inference. Reinforcement learning aims to optimize the reward function and adaptively select the optimal medication regimen. Specific methods include Q-learning (based on the quality function), A-learning (based on the dominance function), and OW-learning (outcome-weighted). In reality, many factors can simultaneously influence medication regimens and patient outcomes. Therefore, this type of technology also needs to be combined with causal inference techniques to overcome the influence of confounding factors on regimen selection. Generally, counterfactual inference methods such as inverse probability weighting and matching are used. This approach requires the assumption of complete collection of confounding factor information and places high demands on the accuracy of counterfactual inference methods; therefore, it is still in the development stage. Furthermore, this type of technology is limited to comparing two or more medication methods and is difficult to adapt to the design of continuous medication regimens such as dosage.

[0007] Overall, existing technologies still have the following limitations in the design of personalized medication regimens: ① It is difficult to address the problem of significant heterogeneity in drug efficacy among patients.

[0008] ② The analysis and comparison of drug efficacy outcomes use targets such as mean and rate. These targets are essentially group indicators and are difficult to adapt to individual differences and individual decision-making needs.

[0009] ③ In an abstract sense, the problem of individualized medication design is to explore the joint distribution P(X,Y|Z) of medication (X) and outcome (Y) given individual characteristics (denoted by Z). Traditional techniques struggle to solve this problem. For example, prediction techniques address the problem of P(Y|Z,X), while reinforcement learning techniques that combine causal inference deal with the difference between P(Y|Z,X) and P(Y0|Z,X0), where Y0 and X0 represent the counterfactual outcome and the counterfactual medication plan, respectively.

[0010] ④ In practice, the evaluation of medication regimens often requires consideration of multiple outcomes, and the data types (such as efficacy and adverse reactions) are diverse. Traditional techniques mostly focus on a single outcome or simplify multiple outcomes into a single objective function.

[0011] According to the inventor, the reasons for the aforementioned limitations include: ① As a new trend, precision medicine has seen a focus of investment in laboratory research, while investment in computer algorithm technology remains limited.

[0012] ② The design of personalized medication regimens needs to consider a large number of patient characteristic variables, which has the characteristics of big data. The data acquisition threshold is high and the mining is difficult, which restricts related innovation work.

[0013] ③ In the field of artificial intelligence technology, predictive and reinforcement learning approaches are relatively prevalent. To some extent, these two existing approaches have created a kind of mental inertia, hindering the establishment and development of new ideas.

[0014] Regarding the aforementioned technologies, where drug efficacy exhibits significant heterogeneity among patients due to the combined influence of multiple factors, existing medication regimen design technologies struggle to guarantee applicability and accuracy, and no effective solution has yet been proposed. Summary of the Invention

[0015] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide a random forest construction method, device, and personalized medication design method to solve the technical problem that existing medication design techniques cannot guarantee applicability and accuracy under the condition that drug efficacy is significantly heterogeneous among patients due to the combined influence of multiple factors. To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution: According to one aspect of the present invention, a method for constructing a random forest for individualized medication regimen design is provided, comprising: When splitting any node in a decision tree, the following strategy is used to reduce the joint distribution heterogeneity of medication regimen variables and patient medication outcome variables within the split node, given individual patient characteristics. Multiple decision trees are built according to the described strategy to generate the random forest.

[0016] Optionally, if both the medication regimen variable and the patient medication outcome variable are continuous or ordinal variables, then reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable within the segmented nodes, given individual patient characteristics, includes: Each variable in the multivariate distribution of the samples to be segmented is ranked and labeled to obtain the labeled rank. The segmentation objective is to maximize the labeled rank distance between the two groups of samples after segmentation into two secondary nodes.

[0017] Optionally, if both the medication regimen variable and the patient medication outcome variable are categorical variables, then reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable within the segmented nodes, given individual patient characteristics, includes: Calculate (n) l +n r )! / (n l ! n r The probability values ​​corresponding to the segmentation results are identified, and the variable corresponding to the highest probability value and its threshold are used as the segmentation basis. Wherein, the n... l and n rThese represent the sample size of the left and right secondary nodes after the segmentation.

[0018] On the other hand, the present invention provides a random forest construction apparatus for individualized medication regimen design, comprising: The decision tree node segmentation unit is used to segment according to the following strategy: to reduce the joint distribution heterogeneity of medication regimen variables and patient medication outcome variables within the segmented node under given individual patient characteristics. The random forest generation unit is used to build multiple decision trees according to the strategy to generate the random forest.

[0019] Optionally, the decision tree node segmentation unit is specifically used to: if both the medication regimen variable and the patient medication outcome variable are continuous variables or ordinal variables, then reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable within the segmented node under given individual patient characteristics includes: Each variable in the multivariate distribution of the samples to be segmented is ranked and labeled to obtain the labeled rank. The segmentation objective is to maximize the labeled rank distance between the two groups of samples after segmentation into two secondary nodes.

[0020] Optionally, the decision tree node segmentation unit is specifically used for: if both the medication regimen variable and the patient medication outcome variable are categorical variables, then reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable within the segmented node under given patient individual characteristics includes: Calculate (n) l +n r )! / (n l ! n r The probability values ​​corresponding to the segmentation results are identified, and the variable corresponding to the highest probability value and its threshold are used as the segmentation basis. Wherein, the n... l and n r These represent the sample size of the left and right secondary nodes after the segmentation.

[0021] According to another aspect of the present invention, a method for designing personalized medication regimens is provided, comprising: Based on the random forest construction method described above, the joint distribution of medication regimen variables and patient medication outcome variables under the condition of individual patient characteristics is estimated. The objectives of the medication regimen are predicted based on the combined distribution; the objectives include one or more of the following: medication experience reference, patient outcome, average efficacy, efficacy percentile, and adverse reaction rate.

[0022] According to another aspect of the present invention, a personalized medication regimen design system is provided, comprising: The aforementioned random forest construction device; The joint distribution estimation unit is used to estimate the joint distribution of medication regimen variables and patient medication outcome variables under the condition of individual patient characteristics, based on the random forest obtained by the random forest construction device. The prediction unit is used to predict the target of the medication regimen based on the joint distribution; the target includes one or more of the following: medication experience patterns, patient outcome distribution, average drug efficacy, drug efficacy percentile, and adverse reaction rate.

[0023] According to another aspect of the present invention, an electronic device is provided, comprising: a processor and a memory; The memory stores a computer-readable program that can be executed by the processor; When the processor executes the computer-readable program, it performs the steps of the method described above.

[0024] According to another aspect of the present invention, a computer-readable storage medium is provided that stores one or more programs, which can be executed by one or more processors to implement the steps in the method described above.

[0025] The random forest construction method, apparatus, and personalized medication regimen design method provided by this invention focus on the changes in multivariate distribution heterogeneity when segmenting any node, minimizing the joint distribution heterogeneity of medication regimen variables and patient medication outcome variables within the segmented node. The construction of its random forest does not depend on any objective or data distribution form, reducing the heterogeneity of efficacy and adverse reactions among patients due to differences in characteristics, and achieving accurate estimation of the joint distribution of medication regimens and patient outcomes. Compared to existing medication regimen design techniques that predict medication outcomes based solely on a given medication regimen and individual conditions, this invention examines the relationship between different medication regimens and patient outcomes (i.e., their joint distribution pattern) under given individual characteristics, using this relationship to formulate an optimal medication strategy tailored to each individual, exhibiting strong individual applicability and high accuracy. Furthermore, compared to reinforcement learning techniques, this invention can handle various types of medication regimens and multiple outcomes (such as continuous, ordinal, and categorical), making its application scope richer and more comprehensive. Attached Figure Description

[0026] Figure 1 Here is a diagram of the structure of the random forest model; Figure 2 This is a flowchart illustrating a random forest construction method provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of a binary distribution rank calibration in the random forest construction method provided in this embodiment of the invention; Figure 4A flowchart illustrating a personalized medication plan design method provided in an embodiment of the present invention; Figure 5 A flowchart illustrating a personalized medication regimen design method according to another embodiment of the present invention; Figure 6 This is a schematic diagram of a random forest construction device provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of a personalized medication plan design device provided in an embodiment of the present invention; Figure 8 This is a structural block diagram of a terminal that can implement the random forest construction method and drug scheme design method of the embodiments of the present invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0028] In related technologies, due to the combined influence of multiple factors, drug efficacy exhibits significant heterogeneity among patients. Existing medication regimen design technologies struggle to guarantee applicability and accuracy, and currently, there is no effective solution to this problem.

[0029] To address the aforementioned problems, this invention proposes a random forest construction method for patient medication regimen design, thereby resolving the technical issues existing in related technologies. A detailed description follows.

[0030] This invention improves upon traditional random forests, making them applicable to multiple outcome scenarios without limiting objectives. Based on this, it proposes a personalized medication design system and method based on conditional joint distribution.

[0031] Random forest is a relatively classic machine learning method. In this invention, it serves as the original prototype of the system and method. Its significance lies in: ① It does not limit the number of explanatory variables or the correlation structure between them. In this invention, the explanatory variables are the patient's genetic characteristics and / or individual dominant characteristics; ② It is applicable to various purposes such as regression and classification, that is, it can analyze different types of outcome variables (including continuous variables, binary variables, survival outcomes, etc.), which are common types of variables for evaluating drug efficacy and adverse reactions; ③ It is compatible with missing values. Since personalized medicine needs to consider a large number of patient characteristics, and in reality, some characteristics of individuals are often incomplete, the feature of being compatible with missing values ​​can make this invention more flexible in application.

[0032] The following section first briefly introduces the basic principles of the random forest algorithm, and then focuses on the improvement method of this invention.

[0033] Random forest, as a type of ensemble algorithm, has the following structure: Figure 1 As shown, the aim is to improve the predictive or classification performance of traditional tree models through a combination approach. Its basic element is a binary tree model. The binary tree model performs multiple splits from top to bottom, each split forming two new secondary nodes. The goal is to maximize the difference in target values ​​(such as mean, rate) between samples at the final nodes under given constraints (such as minimum sample size at the final node). After building a large number of decision trees according to the above rules, a random forest is formed. The "randomness" of the random forest is mainly reflected in two aspects: first, it randomly samples the samples when constructing each binary tree; second, it randomly selects the candidate variable pool at each level node split. Based on this, the random forest constructs tree models in parallel using bootstrap sampling and further ensures the generalization performance of the results using guided aggregation techniques. In the random forest, explanatory variables more relevant to the outcome distribution and training samples closer to the test data have greater weights, thus it can be regarded as an adaptive nearest neighbor method.

[0034] As can be seen from the above principles, the selection of node variables and the binary threshold (i.e., the splitting rule) is crucial for constructing a random forest. For regression analysis, the loss function is typically defined using the mean squared error. (1) The segmentation objective is to minimize the following weighted loss function. (2) in, and These represent the segmentation variable and the segmentation threshold, respectively. and These are the left and right secondary node sample sets after the segmentation, respectively. and These represent the sample size of the left and right secondary nodes after the segmentation.

[0035] For classification analysis, segmentation rules can be based on metrics such as the Gini coefficient or information gain. Taking the Gini coefficient as an example... (3) in, Let Z be the frequency of different classification attribute outcomes in the sample set Z. Similar to Equation (2), the segmentation objective is to minimize the weighted Gini function.

[0036] As can be seen from the above methods, the construction of random forest aims to optimize the "purity" of samples within the same node. Correspondingly, the decomposition of "heterogeneous" data can be achieved by maximizing the differences in inter-group objectives (such as mean and rate).

[0037] This invention focuses on the ability of random forests to decompose heterogeneous data, arguing that it is relatively well-suited to the requirements of personalized medicine treatment design. However, to adapt it to the problem of personalized medicine treatment design, the following improvements are needed: Traditional random forests are a type of prediction (or classification) technique. Based on this approach, developers of personalized medicine regimens typically use both the medication regimen variable / vector (denoted by X) and the individual feature vector (denoted by Z) as explanatory factors to predict the efficacy or adverse reaction outcome (uniformly denoted by Y) of a patient after adopting the regimen, i.e., estimating P(Y|X,Z). This cannot answer the question of what would happen if other medication regimens were chosen given the current combination of patient characteristics, thus making it impossible to compare and select the optimal regimen. An interesting phenomenon is that the patient's individual characteristics also significantly influence the choice of medication regimen; therefore, medication regimen X can be considered a special type of outcome. This perspective differs from traditional prediction approaches; the estimation goal is P(Y,X|Z), that is, the joint distribution of medication regimen X and patient outcome Y given patient characteristics. This invention proposes a personalized medicine regimen design technique adapted to this type of problem.

[0038] Within the technical framework of this invention, both medication regimen variables and outcome variables can be at least one, represented by vectors X and Y respectively. The improvement of this invention over traditional random forests in medication regimen prediction focuses on the setting of the splitting rules, characterized by: ① consideration of multiple outputs (at least one medication regimen and one outcome); ② no need to specify an objective function (mean, etc.). The technical approach is to focus on the changes in multivariate distribution heterogeneity when splitting any node, minimizing the joint distribution heterogeneity of X and Y within the split node.

[0039] In one embodiment, the present invention provides a method for constructing a random forest, combined with... Figure 2 It includes the following steps: S101, When splitting any node of the decision tree, the splitting is performed according to the following strategy: reducing the joint distribution heterogeneity of medication regimen variables and patient medication outcome variables within the split node under the given individual patient characteristics. S103, Build multiple decision trees according to the strategy to generate the random forest.

[0040] In one embodiment, consider the case where both X and Y consist of continuous variables (or ordinal variables). In medication regimen design, continuous variables include medication experience references and average drug efficacy. Since segmentation only involves two secondary nodes, a multivariate test method for any two samples can be considered. As an example, we present a two-sample test method based on multivariate rank. Similar to the univariate case, the joint vector of the samples from the nodes to be segmented can be used. Each component is ranked and labeled within the 0-1 interval (combined with...). Figure 3 (part a) This represents the rank of a sample i after being labeled on the j-th component. The bivariate distribution is a special case; for multivariate rank labeling, see [link to documentation]. Figure 3 Part b of this method can be extended to higher-dimensional states.

[0041] The objective of the partition is to maximize the following function. (4) in, and Define the same formula (2). Let represent the labeled average rank vectors of the left and right secondary node samples, respectively. This formula indicates that the labeled rank distance between the two groups of samples is maximized after splitting into two secondary nodes. Thus, the heterogeneity of the joint distribution after splitting is minimized to the greatest extent.

[0042] In another embodiment, we consider the case where both X and Y are categorical variables, which is very common when studying multiple adverse reactions. Existing techniques often transform different adverse reactions into a single binary outcome (any occurrence is defined as an occurrence, or different weights are artificially assigned), which ignores the high correlation between adverse reactions (one adverse reaction often triggers a series of other reactions). Furthermore, within the technical framework of this invention, it is impossible to transform medication X and traditional outcome Y into a single binary outcome. Therefore, for random forest construction, methods for multivariate categorical variable outcomes must be considered. As mentioned earlier, since the segmentation only involves two secondary nodes, any two-sample multivariate test method can be considered. As an example, we present a joint probability distribution method. According to the hypergeometric distribution, the probability is pre-calculated... Under each segmentation result, identify the probability value corresponding to the variable with the highest probability value. and its threshold This serves as the basis for partitioning. The advantage of this approach is that it avoids traversing all possible partitioning schemes, greatly reducing the computational burden on computer processing units caused by repeated partitioning during the construction of a random forest; at the same time, it ensures that the partitioning point corresponding to the maximum probability value minimizes the heterogeneity of the joint distribution within secondary nodes.

[0043] The construction of the aforementioned random forest does not depend on any specific target or data distribution, which can reduce the heterogeneity of drug efficacy and adverse reactions caused by differences in patient characteristics, and achieve accurate estimation of the joint distribution of medication regimens and patient outcomes. Based on this, random forests can be used to provide estimates for any target of interest, thereby enabling the design of personalized medication regimens.

[0044] Based on this, embodiments of the present invention provide a method for designing personalized medication regimens, combined with Figure 4 Based on the above steps S101 and S103, the following further steps are performed: S105, Based on random forest, estimate the joint distribution of medication regimen variables and patient medication outcome variables under the condition of individual patient characteristics; S107, predict the target of the medication regimen based on the combined distribution; the target includes one or more of the following: medication experience patterns, patient outcome distribution, average drug efficacy, drug efficacy percentile, and adverse reaction rate.

[0045] As an example, following the above method, multiple decision trees (represented as Ntrees) are constructed to... Let represent the data point in the constructed m-th tree that shares the same terminal node as the individual with feature z. Then, for all decision trees with feature z... The average weight of an individual is (5) This makes the weights and features assigned to a training point equal to... The number of trees where individuals reside at the same terminal node is directly proportional to the number of trees in the tree; that is, in the presence of heterogeneity, priority is given to trees based on features. The joint distribution of an individual is estimated from the training samples of its nearest neighbors.

[0046] For a given individual ID, its characteristics are... Treatment regimens and outcomes under certain conditions The joint distribution is estimated as follows (6) in for The density function value. The cumulative distribution function of the joint distribution is... (7) Therefore, it is possible to estimate any target of interest and realize the function of designing multiple individualized medication regimens, as shown in the following example.

[0047] (a) Empirical patterns of medication use: The goal is or This refers to estimating the marginal cumulative distribution or marginal probability distribution of medication regimens (continuous or ordinal variable data) given individual characteristics. These objectives reflect the patterns of medication recommendations made by clinicians for patients with similar characteristics, providing empirical references for both doctors and patients.

[0048] (b) Distribution of patient outcomes: The target is or This refers to the estimation of the marginal cumulative distribution of drug efficacy (continuous or ordinal variable data) or the marginal probability distribution of adverse reactions (categorical variable data) given individual characteristics. These targets reflect the distribution of patient outcomes and can provide doctors and patients with psychological expectations of the outcome.

[0049] (c) Average efficacy estimation: The objective is... and This refers to the expected value of drug efficacy (continuous variable data) given individual characteristics and the expected value under further given medication regimen conditions. By comparing the two, doctors and patients can be informed of the increase or decrease in the expected level of outcome after adopting a certain medication regimen; by comparing the expected values ​​under different medication regimen conditions, the optimal regimen can be identified.

[0050] (d) Percentile estimation of efficacy: The target is and ,in The percentile function can be represented by 50% (for median comparison), 10%, 25%, 75%, 90%, etc. Its application is the same as (c).

[0051] (e) Adverse reaction rate analysis: The target is This refers to the probability estimation of adverse reactions (categorical variable data) given individual characteristics and dosing regimens. It can be used to compare the differences between different dosing regimens (categorical data) for a particular individual to identify the optimal regimen.

[0052] (f) Associated outcome assessment or risk assessment: Conditional joint distribution can also be used to estimate the distribution or probability of distribution of unobserved outcomes of interest (continuous or ordinal variable data) given individual characteristics, medication regimen and some observed outcomes, especially suitable for situations where the time interval between short-term and long-term outcomes is large.

[0053] As an example, the present invention provides a personalized medication plan design method, combined with Figure 5It includes the following steps: N tree models are constructed in parallel. In the construction of each tree model, the training samples are first randomly and repeatedly sampled, followed by random sampling of individual feature variables. During node segmentation, a strategy is employed to minimize the heterogeneity of the joint distribution of X and Y within each segmented node. Specific strategies are described in the previous embodiment and will not be repeated here. Segmentation is repeated until the constraints are met, resulting in the corresponding tree model. The N tree models are ultimately combined to form a random forest. Based on this random forest, a weight function is obtained to estimate the joint distribution of medication regimens and outcomes under individual feature conditions. This function is further used to estimate one or more of the following: medication experience patterns, patient outcome distribution, average drug efficacy, drug efficacy percentiles, and adverse reaction rates.

[0054] As mentioned above, the problem solved by using predictive technology in medication design is merely the prediction of Y given X and Z. However, this embodiment examines the relationship between different treatment plans X and patient outcomes Y (i.e., the joint distribution pattern of the two) given individual characteristics Z. By using this relationship, an optimal medication strategy tailored to each individual can be developed, which has better individual applicability and accuracy.

[0055] With a few exceptions, the effect settings for each predictor in these predictive techniques are based on the "homogeneity" of the population rather than heterogeneity. This leads to issues with the applicability and accuracy of these techniques. This invention, however, is based on and improves upon the random forest method. By focusing on the changes in multivariate distribution heterogeneity during node segmentation, it reduces the heterogeneity of drug efficacy and adverse reactions among patients due to differences in characteristics, thereby improving the individual applicability and accuracy of medication regimens designed based on this random forest model.

[0056] Furthermore, compared to reinforcement learning techniques, this invention can also address the applicability issues of various types of medication regimens (i.e., continuous and categorical X) and multiple outcomes (multiple Y), making its applicability to medication regimen types and outcome categories more abundant and comprehensive.

[0057] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0058] For example, in some embodiments, a computer program can be developed according to the method flow described in the above embodiments and packaged into a computer software product. In this software product, the method flow during computer program execution may further include: receiving patient individual characteristic conditions input by the user, performing calculations according to the method described in the previous embodiments, and outputting the calculation results and corresponding medication regimen recommendations. During this process, computer computing resources can be dynamically allocated based on the real-time requirements of the calculation results and the number of simultaneous calculation tasks. For example, in clinical practice, if it is necessary to provide design opinions for individualized medication regimens for patients in real time, more computing resources need to be allocated to ensure that results can be output at a faster calculation speed. If a large number of patients require medication regimen design within a certain time period, more computing resources can be allocated to the software system to ensure that it can process multiple medication regimen design tasks in parallel.

[0059] In some embodiments, the patient's individual characteristics, medication regimen, and medication outcome data can also be stored in the computer's storage unit, and the algorithm can be continuously improved to enhance accuracy through data accumulation and iteration.

[0060] According to embodiments of the present invention, a random forest construction device for individualized medication regimen design is provided, combining... Figure 6 ,include: The decision tree node segmentation unit 21 is used to segment according to the following strategy: to reduce the joint distribution heterogeneity of medication regimen variables and patient medication outcome variables within the segmented node under given individual patient characteristics. Random forest generation unit 23 is used to build multiple decision trees according to the strategy to generate the random forest.

[0061] Furthermore, embodiments of the present invention also provide a personalized medication plan design system, combined with Figure 7 ,include: The aforementioned random forest construction device (i.e., including decision tree node splitting unit 21 and random forest generation unit 23); The joint distribution estimation unit 25 is used to estimate the joint distribution of medication regimen variables and patient medication outcome variables under the condition of individual patient characteristics based on the random forest obtained by the random forest construction device. The prediction unit 27 is used to predict the target of the medication regimen based on the joint distribution; the target includes one or more of the following: medication experience patterns, patient outcome distribution, average drug efficacy, drug efficacy percentile, and adverse reaction rate.

[0062] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0063] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of the device, can run in a corresponding hardware environment, and can be implemented through software or hardware, wherein the hardware environment includes a network environment.

[0064] Figure 8 This is a structural block diagram of a terminal according to an embodiment of this application, such as... Figure 8 As shown, the terminal may include: one or more (only one is shown) processors 101, memory 103, and transmission devices 105, such as... Figure 8 As shown, the terminal may also include input / output devices 107.

[0065] The memory 103 can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in this embodiment. The processor 101 executes various functional applications and data processing by running the software programs and modules stored in the memory 103, thereby implementing the above-described methods. The memory 103 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 103 may further include memory remotely located relative to the processor 101, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0066] The aforementioned transmission device 105 is used to receive or send data via a network, and can also be used for data transfer between the processor and memory. Specific examples of the network described above may include wired networks and wireless networks. In one example, the transmission device 105 includes a Network Interface Controller (NIC), which can be connected to other network devices and routers via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 105 is a radio frequency (RF) module, used for wireless communication with the Internet.

[0067] Specifically, memory 103 is used to store application programs.

[0068] The processor 101 can call the application stored in the memory 103 through the transmission device 105 to perform the following steps: when splitting any node of the decision tree, the splitting is performed according to the following strategy: reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable under the given individual patient characteristics within the split node; and establishing multiple decision trees according to the strategy to generate the random forest.

[0069] Processor 101 is also configured to perform the following steps: estimating the joint distribution of medication regimen variables and patient medication outcome variables under the condition of individual patient characteristics based on random forest; The objectives of the medication regimen are predicted based on the combined distribution; the objectives include one or more of the following: medication experience reference, patient outcome, average efficacy, efficacy percentile, and adverse reaction rate.

[0070] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0071] Those skilled in the art will understand that the structure of the terminal described above is merely illustrative, and the terminal can be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile internet device (MID), a PAD, or other terminal devices. Figure 8 This does not limit the structure of the aforementioned electronic device. For example, the terminal may also include components that are more... Figure 8 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 8 The different configurations shown.

[0072] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0073] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to execute program code for the above-described method.

[0074] Optionally, in this embodiment, the storage medium may be located on at least one of the network devices in the network shown in the above embodiment.

[0075] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: when splitting any node of a decision tree, the splitting is performed according to the following strategy: reducing the joint distribution heterogeneity of medication regimen variables and patient medication outcome variables under given individual patient characteristics within the split node; and establishing multiple decision trees according to the strategy to generate the random forest.

[0076] Optionally, the storage medium is also configured to store program code for performing the following steps: estimating the joint distribution of medication regimen variables and patient medication outcome variables under conditions of individual patient characteristics based on a random forest; The objectives of the medication regimen are predicted based on the combined distribution; the objectives include one or more of the following: medication experience patterns, patient outcome distribution, average efficacy, efficacy percentile, and adverse reaction rate.

[0077] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0078] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0079] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0080] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0081] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0082] In the embodiments provided in this application, the described device embodiments are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces, indirect couplings, or communication connections between units or modules, and may be electrical or other forms.

[0083] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0084] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0085] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for constructing a random forest for individualized medication regimen design, characterized in that, include: When splitting any node in a decision tree, the following strategy is used to reduce the joint distribution heterogeneity of medication regimen variables and patient medication outcome variables within the split node, given individual patient characteristics. Multiple decision trees are built according to the described strategy to generate the random forest; If both the medication regimen variable and the patient medication outcome variable are continuous or ordinal variables, then reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable within the segmented nodes, given individual patient characteristics, includes: Each variable in the multivariate distribution of the samples to be segmented is ranked and labeled to obtain the labeled rank. The segmentation objective is to maximize the labeled rank distance between the two groups of samples after being segmented into two secondary nodes. If both the medication regimen variable and the patient medication outcome variable are categorical variables, then reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable within the segmented nodes, given individual patient characteristics, includes: Calculate (n) l +n r )! / (n l ! n r The probability values ​​corresponding to the segmentation results are identified, and the variable corresponding to the highest probability value and its threshold are used as the segmentation basis. Wherein, the n... l and n r These represent the sample size of the left and right secondary nodes after the segmentation.

2. A random forest construction device for individualized medication regimen design, characterized in that, include: The decision tree node segmentation unit is used to segment according to the following strategy: to reduce the joint distribution heterogeneity of medication regimen variables and patient medication outcome variables within the segmented node under given individual patient characteristics. A random forest generation unit is used to build multiple decision trees according to the strategy to generate the random forest; The decision tree node segmentation unit is specifically used for: if both the medication regimen variable and the patient medication outcome variable are continuous variables or ordinal variables, then reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable within the segmented node under given individual patient characteristics includes: Each variable in the multivariate distribution of the samples to be segmented is ranked and labeled to obtain the labeled rank. The segmentation objective is to maximize the labeled rank distance between the two groups of samples after being segmented into two secondary nodes. The decision tree node segmentation unit is specifically used for: if both the medication regimen variable and the patient medication outcome variable are categorical variables, then reducing the joint distribution heterogeneity of the medication regimen variable and the patient medication outcome variable within the segmented node under given individual patient characteristics includes: Calculate (n) l +n r )! / (n l ! n r The probability values ​​corresponding to the segmentation results are identified, and the variable corresponding to the highest probability value and its threshold are used as the segmentation basis. Wherein, the n... l and n r These represent the sample size of the left and right secondary nodes after the segmentation.

3. A personalized medication plan design system, characterized in that, include: The random forest construction apparatus according to claim 2; The joint distribution estimation unit is used to estimate the joint distribution of medication regimen variables and patient medication outcome variables under the condition of individual patient characteristics, based on the random forest obtained by the random forest construction device. The prediction unit is used to predict the target of the medication regimen based on the joint distribution; the target includes one or more of the following: medication experience patterns, patient outcome distribution, average drug efficacy, drug efficacy percentile, and adverse reaction rate.

4. An electronic device, characterized in that, include: Processor and memory; The memory stores a computer-readable program that can be executed by the processor; When the processor executes the computer-readable program, it implements the steps of the method as described in claim 1.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs that can be executed by one or more processors to perform the steps of the method as described in claim 1.

Images