A parkinson's disease drug individualization dynamic decision method, system and storage medium

Abstract

This invention relates to a personalized dynamic decision-making method, system, and storage medium for Parkinson's disease medication. It acquires observation data and disease progression information from the patient's current observation window; calculates the current posterior state of the current cycle based on the observation data and disease progression information; matches candidate actions with the patient's current posterior state based on a preset finite set of medication actions; performs hard constraint screening on the candidate actions based on preset safety gating variables; calculates the utility of the candidate actions within the candidate action set based on preset patient preference variables; ranks the candidate actions within the candidate action set based on the calculation results; and finally, filters the candidate action set according to the sequence of each candidate action to obtain and output recommended actions. This not only improves individual matching and enhances patient value alignment but also dynamically adapts to disease progression and supports long-term management.

Description

Technical Field

[0001] This application relates to the field of intelligent decision-making technology, specifically to a personalized dynamic decision-making method, system, and storage medium for Parkinson's disease medication. Background Technology

[0002] Parkinson's disease medication is not a one-time, fixed approach; rather, it requires continuous adjustment as the disease progresses. In real clinical practice, physicians need to make ongoing judgments based on a defined set of treatment actions, including maintaining the current regimen, increasing or decreasing the main medication, adjusting the timing and frequency of administration, adding or withdrawing adjuvant medications, implementing medication switching, and triggering advanced therapy assessments at appropriate times. These decisions are influenced by a combination of factors, including motor symptom control, motor complications, non-motor side effects, patient tolerability, adherence, medication burden, and patient preferences, exhibiting typical sequentiality, individual variability, and safety constraints.

[0003] In existing technologies, Parkinson's disease medication adjustment is a static or quasi-static outpatient decision support system, which typically provides recommendations on whether to adjust medication based on a single outpatient scale, existing prescriptions, and expert rules. Such systems can improve information processing efficiency, but they are difficult to reflect the dynamic evolution of patient status between outpatient visits. Summary of the Invention

[0004] In view of the technical problems existing in the background art, this application provides a personalized dynamic decision-making method, system and storage medium for Parkinson's disease medication. The personalized dynamic decision-making method for Parkinson's disease medication uses a posterior state update driven by "current observation data + previous cycle action / result" as the main axis of state; a finite set of drug action templates as the output boundary; hard constraint screening of safety gating variables as a ranking prerequisite; ranking of candidate actions based on patient preferences and multi-objective utility as a mechanism for distinguishing candidate actions; and a low-confidence conservative branch as a necessary closing point before output, finally outputting recommended actions or conservative actions for physician review.

[0005] In a first aspect, embodiments of this application provide a personalized dynamic decision-making method for Parkinson's disease medication, characterized by specifically including the following steps:

[0006] Acquire patient observation data and disease course information, wherein the disease course information includes the previous cycle's posterior state, actions, and the results after the actions were performed;

[0007] The current posterior state of the current cycle is calculated based on the observed data and the disease course information;

[0008] Candidate actions are matched with the patient's current posterior state based on a preset set of limited drug actions.

[0009] The candidate actions are hard-constrained and filtered according to preset safety gating variables to obtain a set of candidate actions;

[0010] The utility of candidate actions in the candidate action set is calculated based on preset patient preference variables, and the candidate actions in the candidate action set are ranked according to the calculation results.

[0011] The candidate action set is filtered based on the sequence of each candidate action within the candidate action set to obtain and output recommended actions.

[0012] In some embodiments, after acquiring the patient's observation data, the observation data is preprocessed, and missing and conflicting information in the observation data is detected.

[0013] In some embodiments, the disease course information further includes patient compliance, symptom changes after the previous cycle of action, and minimum review interval. The observation data is judged to have low confidence based on the missing information, conflicting information, patient compliance, symptom changes after the previous cycle of action, and minimum review interval. When the observation data meets the low confidence judgment, only a preset conservative action is output. The conservative action includes suspending the adjustment of the plan and performing manual review.

[0014] In some embodiments, the current posterior state includes:

[0015] Motor control status is used to characterize the patient's current level of motor symptom control and recent trends.

[0016] The status of sports complications is used to characterize the treatment-related sports complications.

[0017] Non-motor and tolerance states are used to characterize non-motor and tolerance levels that may influence action selection.

[0018] Treatment exposure and response status is used to characterize the relationship between a patient's current and recent treatment exposure patterns and action-outcomes.

[0019] Disease stage and strategy context are used to characterize the overall context in which the current medication decision is made.

[0020] Secondly, embodiments of this application provide a personalized dynamic decision-making system for Parkinson's disease medication, including:

[0021] The data acquisition module is used to acquire the patient's observation data and disease progress information, which includes the previous cycle's posterior state, actions, and the results after the actions were performed.

[0022] The posterior state update module is used to calculate the current posterior state of the current cycle based on the observation data and the disease course information.

[0023] A finite action template generation module is used to match candidate actions that match the patient's current posterior state based on a preset finite set of drug actions.

[0024] The hard constraint filtering module is used to perform hard constraint filtering on the candidate actions according to preset safety gating variables to obtain a set of candidate actions.

[0025] The utility assessment and ranking module is used to calculate the utility of candidate actions in the candidate action set based on preset patient preference variables, and to rank the candidate actions in the candidate action set according to the calculation results.

[0026] The recommendation output module is used to filter the candidate action set according to the sequence of each candidate action in the candidate action set and output recommended actions.

[0027] In some embodiments, a preprocessing module is further included for preprocessing the observation data and detecting missing and conflicting information in the observation data.

[0028] In some embodiments, a low-confidence conservative branch module is also included. The disease course information further includes patient compliance, symptom changes after the previous cycle's actions, and minimum review interval. The low-confidence conservative branch module performs a low-confidence determination on the observation data based on the missing information, the conflicting information, patient compliance, symptom changes after the previous cycle's actions, and minimum review interval. When the observation data meets the low-confidence determination, only a preset conservative action is output. The conservative action includes temporarily suspending the adjustment of the treatment plan and performing manual review.

[0029] In some embodiments, the current posterior state includes:

[0030] Motor control status is used to characterize the patient's current level of motor symptom control and recent trends.

[0031] The status of sports complications is used to characterize the treatment-related sports complications.

[0032] Non-motor and tolerance states are used to characterize non-motor and tolerance levels that may influence action selection.

[0033] Treatment exposure and response status is used to characterize the relationship between a patient's current and recent treatment exposure patterns and action-outcomes.

[0034] Disease stage and strategy context are used to characterize the overall context in which the current medication decision is made.

[0035] Thirdly, embodiments of this application provide a computer-readable storage medium, which is a non-volatile or non-transient storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the Parkinson's disease drug individualized dynamic decision-making method described above.

[0036] Fourthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the Parkinson's disease drug individualized dynamic decision-making method described in any of the above claims.

[0037] Beneficial Effects: This invention provides a personalized dynamic decision-making method, system, and storage medium for Parkinson's disease medication. It acquires observation data from the patient's current observation window, along with disease progression information including the previous cycle's posterior state, actions, and the results of those actions. Based on the observation data and disease progression information, it calculates the current posterior state of the current cycle using rule-enhanced models, probabilistic models, state-space models, Bayesian update models, or other models that reflect state recursion. It then matches candidate actions to the patient's current posterior state based on a pre-defined finite set of medication actions. The candidate actions are then subjected to hard constraint screening based on pre-defined safety gating variables to obtain a candidate action set. Utility calculations are performed on the candidate actions within the candidate action set based on pre-defined patient preference variables, and the candidate actions are ranked according to the calculation results. Finally, the candidate action set is filtered based on the sequence of each candidate action to obtain and output recommended actions. In this embodiment, the probability of the patient's true physiological / conditional state, which cannot be directly measured, is estimated through the posterior state. The output boundary of the system is defined by a predefined finite set of drug actions. Safety gating variables ensure that the recommended actions do not violate absolute safety rules. Patient preference variables quantify the relative importance that patients attach to different treatment outcomes from multiple dimensions. Finally, a weighted linear summation is performed using a utility function to combine the expected performance of the actions in each dimension with the patient preference weights to obtain personalized utility value. This allows the output recommended actions to not only improve individual matching and enhance patient value alignment, but also to dynamically adapt to disease progression and support long-term management.

[0038] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of this application, the accompanying drawings used in this application will be briefly described below. Obviously, the drawings described below are merely some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort.

[0040] Figure 1 This is a flowchart of a personalized dynamic decision-making method for Parkinson's disease medication in an embodiment of this application;

[0041] Figure 2 This is a flowchart of the action output stage in the embodiments of this application;

[0042] Figure 3 This is a framework diagram of a personalized dynamic decision-making system for Parkinson's disease medication in an embodiment of this application. Detailed Implementation

[0043] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0044] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0045] In this document, the term "comprising" indicates the presence of a described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," with exclusions being otherwise specifically emphasized. Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying one or more of the feature. In the description of embodiments of this application, unless otherwise stated, "a plurality of" means two or more.

[0046] In this text, the term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " in this text generally indicates that the preceding and following related objects have an "or" relationship.

[0047] Parkinson's disease medication is not a one-time, fixed approach; rather, it requires continuous adjustment as the disease progresses. In real clinical practice, physicians need to make ongoing judgments based on a defined set of treatment actions, including maintaining the current regimen, increasing or decreasing the main medication, adjusting the timing and frequency of administration, adding or withdrawing adjuvant medications, implementing medication switching, and triggering advanced therapy assessments at appropriate times. These decisions are influenced by a combination of factors, including motor symptom control, motor complications, non-motor side effects, patient tolerability, adherence, medication burden, and patient preferences, exhibiting typical sequentiality, individual variability, and safety constraints.

[0048] In existing technologies, Parkinson's disease medication adjustment is a static or quasi-static outpatient decision support system, which typically provides recommendations on whether to adjust medication based on a single outpatient scale, existing prescriptions, and expert rules. Such systems can improve information processing efficiency, but they are difficult to reflect the dynamic evolution of patient status between outpatient visits.

[0049] To address the technical problem that existing Parkinson's disease medication adjustments are typically based on single-outpatient scales, existing prescriptions, and expert rules, which, while improving information processing efficiency, fails to reflect the dynamic evolution of patient states across outpatient visits, this application provides a personalized dynamic decision-making method, system, and storage medium for Parkinson's disease medication. This personalized dynamic decision-making method uses a posterior state update driven by both current observation data and previous period actions / outcomes as the main axis; a finite set of medication action templates as the output boundary; hard constraint screening using safety-gated variables as a priori ranking; patient preference and multi-objective utility ranking as a candidate action differentiation mechanism; and a low-confidence conservative branch as a necessary pre-output closure, ultimately outputting recommended or conservative actions for physician review. This achieves improved overall information efficiency while reflecting the dynamic evolution of patient states across outpatient visits, providing patients with better treatment options.

[0050] Please refer to Figure 1 The above is a flowchart illustrating a personalized dynamic decision-making method for Parkinson's disease medication provided in the embodiments of this application, specifically including the following steps:

[0051] S1. Obtain patient observation data and disease course information, which includes the posterior state, actions, and results of the actions performed in the previous cycle.

[0052] Observational data refers to input information within the current observation window that reflects the patient's recent condition, treatment exposure, and related changes. It mainly includes the following categories:

[0053] (1) Structured observation in outpatient clinics

[0054] This includes, but is not limited to, the symptoms stratification, signs stratification, motor fluctuations, motor complications, non-motor symptoms, adverse reaction records, and key points of the current medication regimen recorded by doctors during outpatient visits.

[0055] (2) Summary of Home Monitoring

[0056] This includes, but is not limited to, summary indicators of recent fluctuations such as tremors, activity, sleep-related, gait-related, or other indicators generated by home devices or home recording systems.

[0057] (3) Medication timestamps and treatment exposure records

[0058] This includes, but is not limited to, patient or device records of medication time, whether medication was taken as planned, missed or delayed doses, and recent drug exposure patterns.

[0059] (4) Patient or caregiver report

[0060] This includes, but is not limited to, symptom diaries, fluctuation diaries, patient reports, caregiver reports, descriptions of functional status, and subjective feedback on drug burden or tolerability.

[0061] (5) Adverse reactions and compliance information

[0062] Including but not limited to drowsiness, orthostatic hypotension, nausea, cognitive fluctuations, hallucinations / psychiatric symptoms, impulse control problems, and recent missed doses, delayed doses, missed doses, or other adherence abnormalities.

[0063] Disease course information refers to the complete record of a patient's disease occurrence, development, diagnosis, treatment, and outcome from the onset of the disease to the present time. In this embodiment, disease course information includes the posterior state, actions, and results of the actions performed in the previous cycle. Specifically, the system retrieves the patient's posterior state, actions, and results of the actions performed in the previous cycle from the historical database. Through this disease course information, it is clear what actions were performed in the previous cycle, such as maintenance, dosage increase, dosage decrease, adjustment timing, adjustment frequency, medication increase, medication decrease, or conservative freezing; whether the action was actually confirmed and performed by the doctor; how changes occurred in motor control, motor complications, non-motor tolerance, compliance, and subjective acceptance in the next observation window after execution; and whether these changes have an interpretable correspondence with the direction of the action.

[0064] S2. Calculate the current posterior state of the current cycle based on the observation data and disease course information.

[0065] In this embodiment, the current posterior state of the current cycle is a clinical posterior state obtained by jointly updating the current observation data and disease progression information (i.e., the posterior state, actions, and results of the actions in the previous cycle). Specifically, rule-enhanced models, probabilistic models, state-space models, Bayesian update models, or other models that can reflect state recursion can be used. These models combine the effects of the previous state and actions with the current observations, using probabilistic methods to correct estimates of the patient's physiological / disease state that cannot be directly measured, thereby obtaining the current posterior state of the current cycle.

[0066] S3. Candidate actions are matched with the patient's current posterior state based on a preset set of limited drug actions.

[0067] Specifically, a finite set of drug actions includes at least:

[0068] D-A0: Maintain the current plan

[0069] D-A1: Adjust single dose (increase direction)

[0070] D-A2: Adjust single dose (reduction direction)

[0071] D-A3a: Adjusting the timing of drug administration

[0072] D-A3b: Adjust dosing frequency

[0073] D-A4: Adding medication

[0074] D-A5: Reduced medication

[0075] D-A8: Delay or conservatively output, and upgrade to manual review.

[0076] It should be noted that in this embodiment, "adjusting single dose", "adjusting dosing time", "adjusting dosing frequency", "adding drug" and "reducing drug" are not completely open infinite parameter spaces, but are limited actions restricted by preset step size, preset time template, preset frequency template, preset candidate drug category or preset reduction template.

[0077] In this embodiment, from a set of predefined, limited treatment actions (drug regimens), those actions that are medically reasonable and potentially effective in relation to the patient's current estimated state (posterior state) are identified as candidate actions for the next step.

[0078] S4. Based on the preset safety gating variables, perform hard constraint screening on the candidate actions to obtain a set of candidate actions.

[0079] Specifically, security gating variables include at least:

[0080] The maximum single increase or decrease step size refers to the maximum allowable increase or decrease in the single dose of a certain drug in one adjustment. Its purpose is to prevent acute side effects (such as severe dyskinesia, psychiatric symptoms) or withdrawal malignant syndrome caused by excessive adjustment.

[0081] The minimum review interval refers to the shortest time (usually measured in days or weeks) that must be waited between two adjustments to the same drug. Its purpose is to avoid adding the drug too quickly due to insufficient evaluation, and to prevent the accumulation of adverse drug reactions or misjudgment of efficacy.

[0082] Prohibited or unsuitable drug combinations refer to combinations of certain drugs that may cause serious adverse reactions or antagonistic effects, and are considered absolute or relative contraindications.

[0083] A cognitive or psychiatric red flag refers to a situation where certain medications would significantly worsen a patient's current cognitive impairment or psychiatric symptoms, necessitating immediate discontinuation of these medications.

[0084] Severe drowsiness or alertness risk refers to excessive daytime sleepiness, sleep attacks, or decreased alertness. Some medications can exacerbate this risk and may pose a danger (such as driving).

[0085] Orthostatic hypotension or other tolerance red flags refer to patients having orthostatic hypotension (OH) or intolerable side effects (such as nausea, edema) to certain medications.

[0086] Discontinuation of adherence or signs of exposure instability refer to a patient’s history of irregular medication use, self-discontinuation of medication, missed doses, or overdose. Some regimens may be dangerous due to poor adherence.

[0087] Other safety conditions that directly limit the boundaries of actions.

[0088] In this embodiment, the safety gating variable is an important basis for judging whether "an action is absolutely safe / feasible for the current patient". Therefore, in this embodiment, candidate actions are subjected to hard constraint screening based on the preset safety gating variable to obtain a set of candidate actions that are medically safe, compliant and executable, ensuring that the recommended actions will not cause unacceptable risks.

[0089] S5. Calculate the utility of candidate actions within the candidate action set based on preset patient preference variables, and sort the candidate actions within the candidate action set according to the calculation results.

[0090] Specifically, in this embodiment, preference variables are used to express the differences in trade-offs between different objectives in patient and clinical decision-making scenarios. Preferably, preference variables include at least one or more of the following:

[0091] Tremor control priority refers to the degree of importance a patient places on controlling tremor (especially resting tremor).

[0092] Off-time reduction priority refers to the degree of importance patients place on reducing the "off-time" (the period when the drug's effect wears off and symptoms reappear).

[0093] Dyskinesia minimization priority refers to the degree of importance a patient places on avoiding or reducing dyskinesia (involuntary choreiform movements).

[0094] Cognitive / mental clarity priority refers to the degree of importance a patient places on maintaining clear thinking, memory, and freedom from hallucinations / delusions.

[0095] Sleepiness avoidance refers to the degree of importance a patient places on avoiding excessive daytime sleepiness or sleep attacks.

[0096] The acceptable burden of medication refers to the patient's tolerance for the complexity of the treatment regimen (number of times of daily medication, the relationship between medication and food intake, and the difficulty of operation).

[0097] Caregiver burden sensitivity refers to the degree to which patients are concerned about the time, energy, and professional skills requirements that treatment places on caregivers (usually family members).

[0098] In this embodiment, patient preference variables (i.e., the weights of patients' emphasis on different treatment outcomes) are utilized. By quantifying the patient's preference weights for dimensions such as tremor control, reduction of off-duty time, avoidance of dyskinesia, cognitive clarity, avoidance of drowsiness, and medication and care burden, a personalized total utility value is calculated for each candidate action. These values ​​are then sorted from highest to lowest to match recommended actions to the patient. The final recommended actions are not only "medically correct" but also most aligned with the patient's own values ​​and life goals, thereby achieving truly "patient-centered" healthcare, improving the individualized matching of treatment plans, and enhancing patient value alignment.

[0099] S6. Filter the candidate action set according to the sequence of each candidate action in the candidate action set and output the recommended action.

[0100] Specifically, after obtaining the personalized total utility value of each candidate action and sorting them from high to low according to the utility value, the system selects the top few (usually the first one) as the output recommended actions based on the sorted sequence of candidate actions (i.e., from best to second best) for doctors to confirm.

[0101] In this embodiment, the observation data of the patient's current observation window and the disease information including the posterior state, actions, and results of the previous cycle are obtained. Based on the observation data and disease information, the current posterior state of the current cycle is calculated using a rule-enhanced model, a probabilistic model, a state-space model, a Bayesian update model, or other models that can reflect the recursive relationship of the state. Candidate actions that match the patient's current posterior state are matched based on a preset finite set of drug actions. The candidate actions are hard-constrained and screened according to preset safety gating variables to obtain a set of candidate actions. The utility of the candidate actions in the set of candidate actions is calculated based on preset patient preference variables, and the candidate actions in the set of candidate actions are sorted according to the calculation results. Finally, the set of candidate actions is screened according to the sequence of each candidate action in the set of candidate actions to obtain and output recommended actions. In this embodiment, the probability of the patient's true physiological / conditional state, which cannot be directly measured, is estimated through the posterior state. The output boundary of the system is defined by a predefined finite set of drug actions. Safety gating variables ensure that the recommended actions do not violate absolute safety rules. Patient preference variables quantify the relative importance that patients attach to different treatment outcomes from multiple dimensions. Finally, a weighted linear summation is performed using a utility function to combine the expected performance of the actions in each dimension with the patient preference weights to obtain personalized utility value. This allows the output recommended actions to not only improve individual matching and enhance patient value alignment, but also to dynamically adapt to disease progression and support long-term management.

[0102] In this embodiment, the preset limited set of drug actions is divided into main action templates, extended action templates, and conservative / exceptional action templates.

[0103] The main action templates include D-A0, D-A1, D-A2, D-A3a, and D-A3b. These actions are closest to the routine adjustment units in outpatient clinics. The system can automatically sort and provide a recommended order after passing the safety screening, but their actual execution still requires confirmation from the doctor.

[0104] Extended action templates include D-A4 and D-A5. These actions involve changes or withdrawals in drug composition, and their risks and interpretation costs are generally higher than those of the primary action templates. The system can generate candidates based on the posterior state and participate in restricted ranking, but they are not suitable for automated execution and must be confirmed by a physician.

[0105] Conservative / Exceptional Action Templates include D-A8. When the system triggers a low-confidence conservative branch or a significant security issue occurs, this action can be output directly, indicating that aggressive adjustments are temporarily suspended, the current plan is frozen, and the level of manual review is increased.

[0106] In this embodiment, the main action template only includes low-risk, routine adjustment actions (such as dosage fine-tuning or frequency increase). The system can automatically sort and recommend these actions, which doctors can confirm with a single click. Extended action templates involve changes or reductions in drug composition (such as switching to a different class of drugs or enabling enteric gel infusion), which carry high risks and interpretation costs, requiring doctors to confirm each item and cannot be automatically executed. Conservative / exceptional action templates directly output "postpone adjustment, freeze protocol, or increase review level" when the system has low confidence or safety anomalies, avoiding the system forcibly recommending any aggressive actions when uncertain. This ensures that high-risk decisions are always led by the doctor, while low-risk routine adjustments can be automated, improving efficiency without sacrificing safety, further solidifying boundaries, strengthening safety controls, and preventing automation from overstepping its limits.

[0107] In this embodiment, while outputting recommended actions, the system can also output multiple alternative actions, corresponding key reasons for each action, actions that are removed, and a summary of the reasons for removing actions. It is understood that by outputting a main recommendation plus alternatives in this embodiment, doctors can choose the more suitable option from the main recommendation and alternatives based on the patient's on-site reaction or their own experience, rather than being forced to accept a single output, thus preserving the doctor's discretion. Simultaneously, the corresponding key reasons for each action help doctors quickly verify the rationality of the recommendation. If the reasons do not match the doctor's judgment, they can be promptly overridden, increasing the accuracy and efficiency of the doctor's judgment. Outputting removed actions along with a summary of the reasons for removal eliminates the "black box" feeling, allowing doctors to see not only what the system recommends but also what it does not recommend and why, improving the system's transparency and explainability.

[0108] It should be noted that in this embodiment, the posterior state, safety gating variable, and preference variable are three independent layers. The posterior state layer expresses the "patient's current clinical state"—that is, a multidimensional estimate of movement, complications, non-movement, treatment exposure, and disease stage. The safety gating layer expresses "which actions cannot or should not be performed at present"—based on hard constraints such as absolute contraindications, minimum intervals, and compliance. The preference variable layer expresses "how to rank among multiple safety candidates"—based on the patient's personal values ​​in weighing efficacy, side effects, and burden.

[0109] In this embodiment, the posterior state, safety gating variables, and preference variables are set as three independent layers. This allows the system to generate corresponding annotated summaries when actions are removed, downgraded in ranking, or triggered by conservative branches, enhancing the system's interpretability and clinical trust. Secondly, the layered approach enforces safety priority: the safety gating is executed before preference calculation and has veto power. Any action that fails to pass the safety gating will not enter the preference ranking, thus avoiding the danger of "patient preferences overriding safety," preventing a confusion between the priorities of safety and preferences, and further ensuring patient safety.

[0110] In some embodiments, after acquiring the patient's observation data, the observation data is preprocessed, and missing and conflicting information in the observation data is detected.

[0111] In this embodiment, after acquiring the observation data, the system also needs to perform a series of preprocessing tasks such as quality control, confirmation marking, conflict detection, time alignment, and summary compression to improve data quality and decision reliability, and enhance the system's robustness and practicality.

[0112] Specifically, in this embodiment, quality control identifies anomalous formats (such as incorrect dates or values ​​outside a reasonable range), incorrect times (such as future timestamps), and obviously unreliable values ​​to prevent dirty data from entering state estimation and avoid the model generating absurd posterior states.

[0113] Missing fields are marked and the missing information is retained instead of being deleted. In this embodiment, the missing structure is retained, and the missing data can be naturally processed by Bayesian filtering. This avoids the reduction of sample size or the introduction of bias due to the deletion of records.

[0114] Conflict detection: Identifying inconsistencies between observations from different sources, such as discrepancies between patient self-reports and caregiver reports, or home summaries and outpatient assessments going in opposite directions.

[0115] Time alignment: Maps the current observation to the same observation window or corresponding evaluation sub-window; generates features with a uniform time scale, enabling the state update model to calculate the posterior state at fixed periods (such as daily or weekly).

[0116] Abstract compression: Extracting observations from high-frequency or long sequences into abstract features that can be used for state updates, without requiring the original long sequences to directly enter the decision axis, thereby reducing dimensionality, preserving clinically relevant signals, and allowing state estimation and utility calculation to focus on decision-related statistics.

[0117] like Figure 2 As shown, Figure 2 This is a flowchart of the action output stage in the embodiments of this application.

[0118] In some embodiments, the disease progression information also includes patient compliance, symptom changes after the previous cycle's actions, and the minimum review interval. Based on missing information, conflicting information, patient compliance, symptom changes after the previous cycle's actions, and the minimum review interval, the observed data is assessed for low confidence. When the observed data meets the low confidence threshold, a low-confidence conservative branch is triggered. When a low-confidence conservative branch is triggered, only preset conservative actions are output; these conservative actions include temporarily suspending treatment adjustments and performing manual review. In this embodiment, when a low-confidence conservative branch is triggered, only preset conservative actions are output, which not only blocks unreliable data-driven routine recommendations but also provides a safety net, effectively ensuring patient safety.

[0119] In this embodiment, when the system detects missing / conflicting key observations, abnormal patient compliance, unexplained recent treatment responses, or failure to meet the minimum review interval, a conservative branch is triggered. Regardless of the score of candidate actions derived from the conventional recommendation process (state estimation → utility ranking), these actions are forcibly overridden and cannot be used as the current recommendation output. This prevents unreliable data-driven conventional recommendations and avoids blind adjustments in cases of poor data quality, low compliance, or insufficient evaluation windows, thus preventing the risk of inappropriate medication addition, withdrawal, or change due to erroneous information. Simultaneously, after triggering the low-confidence conservative branch, only conservative actions suitable for safety fallback are output, further ensuring patient safety.

[0120] In this embodiment, the low-confidence conservative branch is not an auxiliary safety hint, but a necessary branch before the final output. Regardless of the current candidate action ranking result, the system should check whether the conservative condition has been triggered before output.

[0121] In this embodiment, the low-confidence conservative branch is triggered by at least one or more of the following conditions:

[0122] (1) Key observations are missing or conflicting

[0123] For example, there are serious gaps in recent medication timestamps, conflicts between outpatient assessments and home summary directions, and significant inconsistencies between patient self-reports and caregiver reports.

[0124] (2) Abnormal compliance

[0125] For example, missed doses, delayed doses, frequent make-up doses, unstable treatment exposure, or inability to confirm the actual medication pattern.

[0126] (3) Recent responses are unexplainable

[0127] For example, the changes in symptoms or side effects after the previous cycle of action are inconsistent with expectations and cannot be reasonably explained by other observations.

[0128] (4) The minimum review interval was not met.

[0129] For example, the last high-impact action just happened recently, and the situation has not yet entered a stable assessment window.

[0130] When a low-confidence conservative branch is triggered, the system must not use the conventional sorting results to output aggressive actions (i.e. recommended actions), but instead output conservative actions such as freezing or postponing the adjustment of the scheme and conducting manual review.

[0131] Specifically, conservative actions, in addition to freezing or suspending the current plan and escalating to manual review, also include limiting the output to D-A0 (maintaining the current plan), listing the key observations that need to be supplemented, explicitly indicating the recommended review time, and covering or masking routine high-impact actions. In this embodiment, when outputting conservative actions, one or more of the above processes can be performed.

[0132] It is important to note that once the conservative branch is triggered, the regular ranking set will no longer be directly used for the recommended output. Even if an action scores highly in the regular utility ranking, it cannot be used as the current recommended action output as long as the conservative branch holds.

[0133] In some embodiments, the current posterior state includes:

[0134] Motor control status is used to characterize the patient's current level of motor symptom control and recent trends.

[0135] The status of sports complications is used to characterize the treatment-related sports complications.

[0136] Non-motor and tolerance states are used to characterize non-motor and tolerance levels that may influence action selection.

[0137] Treatment exposure and response status is used to characterize the relationship between a patient's current and recent treatment exposure patterns and action-outcomes.

[0138] Disease stage and strategy context are used to characterize the overall context in which the current medication decision is made.

[0139] Specifically, in this embodiment, the motor control state is used to characterize the patient's current level of motor symptom control and recent trends, including but not limited to the following: vibration burden, bradykinesia burden, rigidity burden, axial / gait / freezing burden, intraday motor fluctuation phase and its trend.

[0140] Exercise complication status is used to characterize treatment-related exercise complication status, including but not limited to OFF burden, wearing-off tendency, dyskinesia burden, ON / OFF variability, morning OFF burden, or other variability issues coupled with treatment exposure.

[0141] Non-motor and tolerance states are used to characterize non-motor and tolerance levels that may affect action choices, including but not limited to drowsiness, fatigue, autonomic discomfort, cognitive fluctuations, vulnerability to mental symptoms, mood changes, and vulnerability to impulse control.

[0142] Treatment exposure and response status is used to characterize the relationship between a patient's current and recent treatment exposure patterns and action-outcomes, including but not limited to current medication composition, recent changes in LEDD, recent dose-time pattern, direction of symptom and side effect response after recent actions, whether the action was actually performed, and whether the changes after performance are interpretable.

[0143] Disease stage and strategy context are used to characterize the overall context in which the current medication decision is made, including but not limited to the disease stage, the current main problem domain, the current treatment goal priority, and whether it is in the window of conservative management.

[0144] It's important to note that the core diagnostic and treatment target for Parkinson's disease is motor symptoms. Without understanding the specifics of tremor, bradykinesia, rigidity, gait, and intraday fluctuations, it's impossible to determine whether current treatment is adequate or to select targeted medications.

[0145] For Parkinson's disease patients undergoing long-term medication (such as levodopa) treatment, motor complications (end-of-dose phenomenon, dyskinesia, and on / off fluctuations) become the main obstacle to their quality of life. Ignoring this aspect might lead the system to recommend increasing the medication dosage, which, while improving off-off time, could induce severe dyskinesia, resulting in more harm than good.

[0146] Furthermore, understanding the patient's actual medication, dosage, adherence, and past response to actions is crucial for establishing a causal chain of "action → result." If any one of these is overlooked, the system may mistakenly attribute "ineffectiveness" to insufficient dosage and blindly increase the dosage, leading to the risk of overdose.

[0147] It is also important to note that the treatment goals and risk tolerance differ significantly between the early, middle, and late stages. The same degree of dyskinesia may be unacceptable in early-stage patients but tolerable in late-stage, disabling OFF patients.

[0148] Therefore, the complexity of Parkinson's disease treatment decisions requires the system to characterize the patient's current state from a multi-dimensional, dynamic, and interconnected perspective. In this embodiment, the posterior state must include five core components: motor control state, motor complication state, non-motor and tolerance state, treatment exposure and response state, and disease stage and strategy context. These five core components respectively cover the therapeutic target (motor), treatment contradictions (complications), safety boundaries (non-motor), causal chain (exposure and response), and decision context (stage and context). The absence of any one of these components will cause the system to lose crucial decision-making basis, leading to potentially ineffective, unsafe, or out-of-touch recommendations.

[0149] Thirdly, embodiments of this application provide a personalized dynamic decision-making system for Parkinson's disease medication.

[0150] like Figure 3 As shown, Figure 3 This is a framework diagram of a personalized dynamic decision-making system for Parkinson's disease medication, as described in an embodiment of this application. The personalized dynamic decision-making system for Parkinson's disease medication includes:

[0151] The data acquisition module is used to acquire the patient's observation data and disease progress information. The disease progress information includes the previous cycle's posterior status, actions, and the results after the actions were performed.

[0152] The posterior state update module is used to calculate the current posterior state of the current cycle based on observation data and disease course information.

[0153] A finite action template generation module is used to match candidate actions based on a preset finite set of drug actions with the patient's current posterior state.

[0154] The hard constraint filtering module is used to perform hard constraint filtering on candidate actions based on preset safety gating variables to obtain a set of candidate actions.

[0155] The utility assessment and ranking module is used to calculate the utility of candidate actions within the candidate action set based on preset patient preference variables, and to rank the candidate actions within the candidate action set according to the calculation results.

[0156] The recommendation output module is used to filter the candidate action set based on the sequence of each candidate action in the candidate action set and output recommended actions.

[0157] In this embodiment, the data acquisition module acquires the observation data of the patient's current observation window and the disease information, which includes the posterior state, actions, and results of the previous cycle, and then sends the observation data and disease information to the posterior state update module. The posterior state update module calculates the current posterior state of the current cycle based on the observation data and disease information using a rule-enhanced model, a probabilistic model, a state-space model, a Bayesian update model, or other models that can reflect the recursive relationship of the state. The finite action template generation module matches candidate actions that match the patient's current posterior state based on a preset finite set of drug actions. The hard constraint screening module performs hard constraint screening on the candidate actions based on preset safety gating variables to obtain a set of candidate actions. The utility evaluation and ranking module performs utility calculations on the candidate actions in the set of candidate actions based on preset patient preference variables and ranks the candidate actions in the set of candidate actions based on the calculation results. Finally, the recommendation output module filters the set of candidate actions based on the sequence of each candidate action in the set of candidate actions and outputs recommended actions. In this embodiment, the probability of the patient's true physiological / conditional state, which cannot be directly measured, is estimated through the posterior state. The output boundary of the system is defined by a predefined finite set of drug actions. Safety gating variables ensure that the recommended actions do not violate absolute safety rules. Patient preference variables quantify the relative importance that patients attach to different treatment outcomes from multiple dimensions. Finally, a weighted linear summation is performed using a utility function to combine the expected performance of the actions in each dimension with the patient preference weights to obtain personalized utility value. This ensures that the output recommended actions not only improve individual matching and enhance patient value alignment, but also dynamically adapt to disease progression and support long-term management.

[0158] In some embodiments, a preprocessing module is also included for preprocessing the observation data and detecting missing and conflicting information in the observation data.

[0159] In some embodiments, a low-confidence conservative branch module is also included. The disease course information also includes patient compliance, symptom changes after the previous cycle's actions, and minimum review interval. The low-confidence conservative branch module performs a low-confidence determination on the observation data based on missing information, conflicting information, patient compliance, symptom changes after the previous cycle's actions, and minimum review interval. When the observation data meets the low-confidence determination, only the preset conservative actions are output. The conservative actions include temporarily suspending the adjustment of the treatment plan and performing manual review.

[0160] In some embodiments, the current posterior state includes:

[0161] Motor control status is used to characterize the patient's current level of motor symptom control and recent trends.

[0162] The status of sports complications is used to characterize the treatment-related sports complications.

[0163] Non-motor and tolerance states are used to characterize non-motor and tolerance levels that may influence action selection.

[0164] Treatment exposure and response status is used to characterize the relationship between a patient's current and recent treatment exposure patterns and action-outcomes.

[0165] Disease stage and strategy context are used to characterize the overall context in which the current medication decision is made.

[0166] Thirdly, embodiments of this application provide a computer-readable storage medium, which is a non-volatile or non-transient storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the Parkinson's disease drug individualized dynamic decision-making method described above.

[0167] Fourthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the Parkinson's disease drug individualized dynamic decision-making method described in any of the above claims.

[0168] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application of the technical solution and the constraints involved. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0169] When the embodiments of this application are implemented using software, they can be implemented entirely or partially in the form of a computer program product. That is, the implementation of all or part of the processes in the methods of the above embodiments can also be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0170] It should be noted that this application is not limited to the above-described embodiments. The above embodiments are merely examples, and any embodiments with the same structure and effect as the technical concept within the scope of this application are included in the technical scope of this application. Furthermore, various modifications that can be conceived by those skilled in the art to the embodiments, and other ways of constructing by combining some of the constituent elements of the embodiments, without departing from the spirit of this application, are also included in the scope of this application.

Claims

1. A personalized dynamic decision-making method for Parkinson's disease medication, characterized in that, Specifically, the following steps are included: Acquire patient observation data and disease course information, wherein the disease course information includes the previous cycle's posterior state, actions, and the results after the actions were performed; The current posterior state of the current cycle is calculated based on the observed data and the disease course information; Candidate actions are matched with the patient's current posterior state based on a preset set of limited drug actions. The candidate actions are hard-constrained and filtered according to preset safety gating variables to obtain a set of candidate actions; The utility of candidate actions in the candidate action set is calculated based on preset patient preference variables, and the candidate actions in the candidate action set are ranked according to the calculation results. The candidate action set is filtered based on the sequence of each candidate action within the candidate action set to obtain and output recommended actions.

2. The personalized dynamic decision-making method for Parkinson's disease medication according to claim 1, characterized in that, After obtaining the patient's observation data, the observation data is preprocessed, and missing and conflicting information in the observation data is detected.

3. The personalized dynamic decision-making method for Parkinson's disease medication according to claim 2, characterized in that, The disease course information also includes patient compliance, symptom changes after the previous cycle of action, and minimum review interval. Based on the missing information, conflicting information, patient compliance, symptom changes after the previous cycle of action, and minimum review interval, the observation data is judged to have low confidence. When the observation data meets the low confidence judgment, only the preset conservative action is output. The conservative action includes temporarily suspending the adjustment of the plan and performing manual review.

4. The personalized dynamic decision-making method for Parkinson's disease medication according to claim 3, characterized in that, The current posterior state includes: Motor control status is used to characterize the patient's current level of motor symptom control and recent trends. The status of sports complications is used to characterize the treatment-related sports complications. Non-motor and tolerance states are used to characterize non-motor and tolerance levels that may influence action selection. Treatment exposure and response status is used to characterize the relationship between a patient's current and recent treatment exposure patterns and action-outcomes. Disease stage and strategy context are used to characterize the overall context in which the current medication decision is made.

5. A personalized dynamic decision-making system for Parkinson's disease medication, characterized in that, include: The data acquisition module is used to acquire the patient's observation data and disease progress information, which includes the previous cycle's posterior state, actions, and the results after the actions were performed. The posterior state update module is used to calculate the current posterior state of the current cycle based on the observation data and the disease course information. A finite action template generation module is used to match candidate actions that match the patient's current posterior state based on a preset finite set of drug actions. The hard constraint filtering module is used to perform hard constraint filtering on the candidate actions according to preset safety gating variables to obtain a set of candidate actions. The utility assessment and ranking module is used to calculate the utility of candidate actions in the candidate action set based on preset patient preference variables, and to rank the candidate actions in the candidate action set according to the calculation results. The recommendation output module is used to filter the candidate action set according to the sequence of each candidate action in the candidate action set and output recommended actions.

6. The personalized dynamic decision-making system for Parkinson's disease medication according to claim 1, characterized in that, It also includes a preprocessing module for preprocessing the observation data and detecting missing and conflicting information in the observation data.

7. The personalized dynamic decision-making system for Parkinson's disease medication according to claim 2, characterized in that, It also includes a low-confidence conservative branch module. The disease course information includes patient compliance, symptom changes after the previous cycle's actions, and minimum review interval. The low-confidence conservative branch module performs a low-confidence determination on the observation data based on the missing information, the conflicting information, patient compliance, symptom changes after the previous cycle's actions, and minimum review interval. When the observation data meets the low-confidence determination, only the preset conservative actions are output. The conservative actions include temporarily suspending the adjustment of the treatment plan and performing manual review.

8. The personalized dynamic decision-making system for Parkinson's disease medication according to claim 3, characterized in that, The current posterior state includes: Motor control status is used to characterize the patient's current level of motor symptom control and recent trends. The status of sports complications is used to characterize the treatment-related sports complications. Non-motor and tolerance states are used to characterize non-motor and tolerance levels that may influence action selection. Treatment exposure and response status is used to characterize the relationship between a patient's current and recent treatment exposure patterns and action-outcomes. Disease stage and strategy context are used to characterize the overall context in which the current medication decision is made.

9. A computer-readable storage medium, wherein the computer-readable storage medium is a non-volatile storage medium or a non-transient storage medium, and a computer program is stored thereon, characterized in that, When the computer program is run by the processor, it executes the steps of the personalized dynamic decision-making method for Parkinson's disease medication as described in any one of claims 1 to 4.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the personalized dynamic decision-making method for Parkinson's disease medication as described in any one of claims 1 to 4.

Images